Files

112 KiB
Raw Permalink Blame History

title, status, date, scope
title status date scope
MAC-address-based footfall sensing in the EU/Germany: legal and privacy research note research-note 2026-11-14 Passive BLE/Wi-Fi MAC sensing in public spaces, with a focus on Heilbronn and the Code for Heilbronn PAX counter design

MAC-address-based footfall sensing in the EU/Germany

Purpose. This note examines the legal and privacy status of passive BLE/Wi-Fi sensing of MAC addresses in public spaces in Germany, and what that means for a Heilbronn-style footfall deployment that aggregates device counts. It is written for a technical paper, not as a legal opinion.

Disclaimer. This is a research note, not legal advice. The decisive German question — whether passive reception of broadcast MAC addresses is an "access" within the meaning of § 25 TDDDG — is not settled by published case law or supervisory guidance. A deployment should obtain a written opinion from German counsel before going live, especially if it is operated by or for a public body.

Evidence labels used below. [Law] binding EU/German law; [CJEU] binding interpretation by the Court of Justice; [Guidance] EDPB, DSK, CNIL, AP or ICO guidance (persuasive, not binding); [DPA] a national supervisory authority's published position or annual report; [Enforcement] an enforcement decision or court judgment; [Technical] code, standards, or peer-reviewed measurement literature. Full references are in §8.

Why the Heilbronn case is not generic. The deployed design described by Code for Heilbronn is a passive BLE scanner that hashes device addresses and keeps them for at most one hour, transmits only counts over LoRaWAN, and publishes 15-minute counts. The upstream reference implementation (ESP32-Paxcounter/libpax) uses a 16-bit value derived from the last two bytes of the MAC address in a 60-second bitmap. These details matter: they determine whether the system is processing personal data, whether it is pseudonymising or anonymising, and what a DPIA must actually assess. See §2.2 and §5.2.


1. Bottom line

  1. In this context, MAC addresses should be treated as personal data unless a documented anonymisation assessment proves otherwise. A MAC address is a device identifier; combined with sensor location and time it singles out a device and can be linked to a person by the controller or by third parties. The CJEU's test is relative but broad: "all the means likely reasonably to be used" by the controller or another person, including means accessible through third parties. [CJEU] Breyer, paras 42–46; [Law] GDPR Art. 4(1), Recital 26. The EDPB's 2026 draft anonymisation guidelines expressly use a MAC address as an example of a hardware identifier that can identify a device and, in turn, an individual. [Guidance] EDPB Guidelines 02/2026, para. 25.

  2. Hashing a MAC address is pseudonymisation, not anonymisation. The operator's statement that "no conclusions can be drawn about devices or persons" is not supported by hashing alone. An unsalted hash of a 48-bit MAC is enumerable; a short or truncated hash (the upstream reference uses 16 bits) has a small space and frequent collisions; a keyed hash can be re-identified by whoever holds the key, because the MAC space is small enough to enumerate. [Guidance] EDPB Guidelines 01/2025 (consultation version), para. 22; [Enforcement] Conseil d'État, 8 Feb 2017, no. 393714, para. 8; [DPA] Autoriteit Persoonsgegevens, "Wifi and bluetooth". Only the aggregated output can become anonymous, and only if the legal test is met.

  3. Edge processing does not remove the need for a legal basis. The sensor necessarily processes the raw MAC in memory to deduplicate. That is processing of personal data. Anonymising the output later does not retroactively make the collection stage lawful. The legal basis must cover the collection and the anonymisation step, not just the published counts. [Guidance] EDPB Guidelines 01/2025, para. 23; [Guidance] CNIL, 24 Feb 2020 (anonymisation within minutes still requires a legal basis, information and opt-out).

  4. ePrivacy Art. 5(3) and § 25 TDDDG may apply independently of the GDPR, and independently of whether the data are personal. Both provisions require prior consent for storing information in, or gaining access to information already stored in, terminal equipment. § 25 TDDDG expressly applies regardless of personal data. [Law] ePrivacy Directive Art. 5(3); [Law] TDDDG § 25(1), (2); [Guidance] DSK, OH Digitale Dienste (Nov 2024), para. 26.

  5. Whether passive sniffing of broadcast MACs is an "access" under Art. 5(3)/§ 25 is the central unresolved question. The better textual reading is arguably that it is not: the sensor does not store anything in the phone, does not instruct the phone to disclose stored information, and does not retrieve data from the phone's memory; it merely receives a frame the phone broadcasts autonomously. German legislative materials speak of "Abruf" and "Auslesen" (retrieval/readout), which imply an active access. [Law] BT-Drs. 19/27441, p. 38. But the EDPB's broad functional approach — MAC addresses are "information" stored in the terminal equipment, and a receiving entity need not be the one instructing the transmission — creates material risk the other way. [Guidance] EDPB Guidelines 2/2023, paras 39, 42. There is no published EU or German decision on passive Wi-Fi/BLE sniffing under Art. 5(3)/§ 25.

  6. If § 25 applies, the project cannot rely on legitimate interest, signage or an opt-out. § 25 requires prior consent meeting GDPR Art. 4(11) and Art. 7; the only exceptions are transmission and strict necessity for a digital service explicitly requested by the user. [Law] TDDDG § 25(2). Consent at city scale is practically impossible: walking past a sensor, leaving Wi-Fi on, or seeing a sign is not consent. [CJEU] Planet49, paras 52–65; [Law] GDPR Recital 32. If § 25 applies, the viable routes are a statutory legal basis or a sensor that does not extract MAC addresses. Separately, if the data are personal, Art. 14 information and Art. 21(4) objection notice are required; if the data are truly anonymous, there is no GDPR notification duty, but transparency remains good practice.

  7. For a public body, Art. 6(1)(f) is unavailable. A municipality acting in the performance of its tasks must rely on Art. 6(1)(e) read with Art. 6(3) and a legal basis in Union or Member State law, or on consent. [Law] GDPR Art. 6(1) last sentence, Art. 6(3). For a private association or company, Art. 6(1)(f) is available but the balancing test is strict for covert public-space tracking; German and Dutch DPAs have repeatedly found that the data subject's interests prevail. [DPA] ULD Schleswig-Holstein, TB 2019, pp. 97–99; [DPA] Autoriteit Persoonsgegevens, "Legal bases for WiFi tracking and Bluetooth tracking"; [CJEU] M5A, KNLTB.

  8. A DPIA is almost certainly mandatory. Continuous sensing of a publicly accessible inner-city area is the paradigm of "systematic monitoring of a publicly accessible area on a large scale" under Art. 35(3)(c). The Baden-Württemberg supervisory authority's private-sector "Muss-Liste" explicitly lists offline tracking of customer movements in department stores/shopping centres and the non-intended use of device sensors or radio signals to determine location or movement over a substantial period. [Law] GDPR Art. 35(3)(c); [Guidance] LfDI BW, Muss-Liste, Nos. 2 and 10. There is no public-sector positive list in Baden-Württemberg, so for the city the statutory criterion applies directly. [Guidance] LfDI BW/MJM BW, Schwellenwertanalyse (2024).

  9. German enforcement has been administrative, not fine-based, for this exact use case. The ULD Schleswig-Holstein investigated two city-marketing operators (a Verein and a GmbH) that passively captured probe requests and concluded there was no legal basis; both stopped. The Bavarian authority inspected a fashion chain's Wi-Fi tracking test after a complaint; the company stopped. The Hamburg authority's 2014/2015 report treats WLAN/Bluetooth indoor tracking as highly intrusive and personal-data processing requiring a legal basis. [DPA] ULD TB 2019, §5.4.8; [DPA] BayLDA TB 2015/16, §§3.10, 22.10; [DPA] HmbBfDI TB 2014/2015, §V.9. No published German fine specifically for passive Wi-Fi/BLE footfall tracking was found; that is not evidence of legality.

  10. The most important EU enforcement case ended in the regulator's loss — on evidence, not on the law. The Dutch DPA fined the municipality of Enschede €600,000 for Wi-Fi tracking in the city centre (pseudonymised MAC + location, used to count unique visitors). The fine was annulled by the Rechtbank Overijssel on 2 February 2024 because the AP had not proved that the municipality processed personal data; the Raad van State upheld that outcome on 29 July 2026, essentially on the procedural ground that the AP's new "direct identification" argument had not been in the original decision. The case is not a safe harbour: it shows that the legal status is fact-specific and that the regulator must prove identifiability. [Enforcement] ECLI:NL:RBOVE:2024:594; ECLI:NL:RVS:2026:4403.

  11. The French JCDecaux case is the strongest warning against relying on hashing/salting. The Conseil d'État upheld the CNIL's refusal of a Wi-Fi tracking experiment on the La Défense esplanade: truncating the MAC, adding a salt and keyed hashing still left the controller able to identify and link records, and the purpose — measuring routes and repeat passages — was itself incompatible with anonymisation. [Enforcement] Conseil d'État, 8 Feb 2017, no. 393714, paras 7–8.

  12. For Heilbronn, the defensible architecture is: no raw MAC retention; passive reception only; per-sensor, per-window keyed hashes; no cross-sensor or cross-day linkage; aggregate to zones/time bins; k-threshold plus suppression and, where useful, differential privacy; a DPIA before deployment; on-site signage; a real opt-out; and Art. 28 contracts. If the project cannot document why § 25 does not apply, it should either obtain consent (which is impractical at city scale) or switch to a sensor that does not extract MAC addresses. The engineering recommendations in §7 are starting points, not legal safe harbours.


2.1 The statutory test

[Law] GDPR Art. 4(1) defines personal data as "any information relating to an identified or identifiable natural person"; an identifiable person is one "who can be identified, directly or indirectly, in particular by reference to an identifier such as a name, an identification number, location data, an online identifier or to one or more factors specific to the physical, physiological, genetic, mental, economic, cultural or social identity of that natural person."

[Law] Recital 26 adds the decisive qualifier: account should be taken of "all the means reasonably likely to be used, such as singling out, either by the controller or by another person, to identify the natural person directly or indirectly." The assessment must consider "all objective factors, such as the costs of and the amount of time required for identification, taking into consideration the available technology at the time of the processing and technological developments." Anonymous information — information that does not relate to an identified or identifiable person, or personal data rendered anonymous so that the data subject is no longer identifiable — falls outside the GDPR.

[Law] Recital 30 explicitly includes online identifiers: "natural persons may be associated with online identifiers provided by their devices, applications, tools and protocols, such as internet protocol addresses, cookie identifiers or other identifiers such as radio frequency identification tags." It warns that this "may leave traces which, combined with unique identifiers and other information received by the servers, may be used to create profiles of the natural persons and identify them."

[CJEU] Breyer (C-582/14) is the leading case. A dynamic IP address can be personal data for a website operator where the operator has the legal means to obtain additional identifying information from the ISP; it is not necessary that all the identifying information be in one person's hands. [CJEU] Breyer, paras 42–46. The "means likely reasonably to be used" test excludes only identification requiring a "disproportionate effort in terms of time, cost and manpower." [CJEU] Breyer, para. 46. The possibility of combining data must not be "purely hypothetical." [CJEU] Breyer, para. 49.

[CJEU] The Court has repeatedly read the concept broadly. In Nowak, it held that "personal data" is not limited to information about private or sensitive matters; exam scripts and examiner comments can be personal data. [CJEU] C-434/16, para. 35. In Vyriausioji tarnybinės etikos komisija, it confirmed that indirect identification can result from combining data. [CJEU] C-184/20. In Gesamtverband Autoteile-Handel, it held that a vehicle identification number can be personal data where it can be linked to an identifiable person; the link can arise from the content, purpose or effect of the information. [CJEU] C-319/22, paras 45–49. In IAB Europe, it held that a TC String — a sequence of characters encoding a user's consent preferences — can be personal data where it can be linked to an identifiable user, even though it does not contain a name. [CJEU] C-604/22. In EDPS v SRB, it clarified that pseudonymised data are not personal data "in all cases and for every person": the assessment is relative to the recipient's reasonably available means of re-identification. [CJEU] C-413/23 P, paras 55–56, 82–85. But the same judgment confirms that pseudonymised data remain personal for the entity that holds the key. [CJEU] C-413/23 P, para. 76.

[Guidance] The Article 29 Working Party's Opinion 4/2007 remains the canonical explanation: "personal data" covers "any information" relating to a person by reason of its content, purpose or result; identification can be indirect; the assessment is contextual and risk-based. [Guidance] WP29 Opinion 4/2007, pp. 9–12.

2.2 MAC addresses in the Heilbronn pipeline

A MAC address is a 48-bit identifier assigned to a network interface. For a universally administered address, the first 24 bits are an Organizationally Unique Identifier (OUI) that identifies the vendor; the last 24 bits are assigned by the vendor. Randomised addresses set the locally administered bit and do not carry a real vendor OUI. It is a device identifier, not a person identifier. But in a footfall deployment it is processed together with:

  • the receiving sensor's identity and location;
  • a timestamp (the Heilbronn data are on a ~15-minute cadence);
  • the fact that the same device was seen before (deduplication);
  • potentially, the signal strength and channel;
  • potentially, the device type inferred from the OUI.

That combination is what makes the data identifying. A single MAC address plus "sensor 7, 14:15" can single out a device and its bearer in a small area; a sequence of such records across sensors can reconstruct a path. The controller may not know the person's name, but it can treat the device bearer differently (count them once, exclude them, measure repeat visits), which is the functional meaning of "identifiable" in the EDPB's 2026 draft. [Guidance] EDPB Guidelines 02/2026, paras 19–25.

[Technical] The Heilbronn deployment, as described by Code for Heilbronn, uses a Heltec Wireless Paper board with an ESP32 (BLE/Wi-Fi) and an SX1262 LoRa transceiver. The sensor "receives these packets and remembers the device addresses. These MAC addresses are hashed and stored for at most one hour" (author's translation). Only the number of counted addresses is transmitted; hashed addresses do not leave the sensor. [Technical] Code for Heilbronn, "Passantenzahlen messen mit LoRa" (18 Sep 2025). The upstream reference implementation (ESP32-Paxcounter/libpax) uses a passive BLE scan (no scan requests) and, in mac_add(), a 16-bit ID built from the last two bytes of the address, stored in a 65,536-bit bitmap that is cleared every 60 seconds. It counts only locally administered (randomised) addresses and ignores universally administered (burned-in) addresses. [Technical] libpax/lib/libpax/libpax.cpp; ESP32-Paxcounter v3.8.0; see also the project's code audit, 09-paxcounter-code-audit.md. The upstream documentation says the opposite (that personal MAC addresses are not evaluated), so the actual deployed firmware must be verified; the code audit flags this discrepancy. The upstream project's own legal note warns that MAC sniffing may be illegal depending on jurisdiction and links to Dutch/EU guidance, a decision, and a German Bundestag analysis. [Technical] Paxcounter legal note.

This design is materially better than storing raw MACs. It is not, by itself, a legal anonymisation guarantee:

  • the raw MAC is observable at the sensor before hashing;
  • the hash length and algorithm of the deployed firmware are undocumented; the upstream 16-bit ID has a small space and frequent collisions;
  • the one-hour cache creates a one-hour linkability window per sensor;
  • if the hash is unsalted, the 48-bit MAC space is small enough to enumerate; if it is keyed, the key holder can re-identify;
  • the published 15-minute per-sensor counts can reveal a single device in low-traffic cells.

[Guidance] The EDPB's 2026 draft says a hardware identifier such as a MAC address "can identify the device and, in turn, the individual using the device." [Guidance] EDPB Guidelines 02/2026, para. 25. The Dutch DPA's position is even more direct: MAC addresses combined with sensor location and time are "indirectly identifying personal data" because they can be combined with camera images, payment data, hotspot logins, or RFID records. [DPA] Autoriteit Persoonsgegevens, "Wifi and bluetooth".

Conclusion on classification. A MAC address is not automatically personal data in every conceivable context, but in a public-space footfall system with location, time and deduplication, it is very likely personal data (or at least pseudonymous personal data). The burden is on the controller to demonstrate that the data are anonymous for every relevant recipient. A claim that "we hash, therefore it is anonymous" is not sufficient. See §5.

2.3 Pseudonymous versus anonymous

[Law] GDPR Art. 4(5) defines pseudonymisation as processing personal data "in such a manner that the personal data can no longer be attributed to a specific data subject without the use of additional information," provided the additional information is kept separately and protected. Pseudonymised data remain personal data for the controller that holds the additional information, and for any recipient that can reasonably re-identify the person; they may be anonymous for a recipient who cannot. [CJEU] C-413/23 P, paras 82–85. [Guidance] EDPB Guidelines 01/2025 (consultation version), para. 22: pseudonymised data that can be attributed to a person by additional information are personal data, "even if pseudonymised data and additional information are not in the hands of the same person"; even if the controller erases the additional information, the data become anonymous only if the conditions for anonymity are met.

[Guidance] The EDPB's 2026 draft sets out the current analytical framework: anonymity is assessed from the perspective of each relevant entity; the test is whether the person can be distinguished from others using means reasonably likely to be used, in a way that makes it possible to treat them differently. [Guidance] EDPB Guidelines 02/2026, executive summary and paras 19–25. The draft's three criteria — No Record Isolation, No Linkage, No Inference — are discussed in §5.1.

[Enforcement] The Conseil d'État's JCDecaux decision is the clearest judicial statement that technical pseudonymisation is not enough. JCDecaux proposed to truncate the last half-octet of the MAC, add a salt, and apply keyed hashing. The court held that these processes "seek to prevent third parties from accessing the data" but "leave the processing manager able to identify the persons concerned and do not prevent correlating records relating to the same individual or inferring information about him" (author's translation). [Enforcement] CE, 8 Feb 2017, no. 393714, para. 8. The court also held that the purpose — measuring the repetition of passages and the routes taken — was incompatible with anonymisation. [Enforcement] Ibid. The Dutch DPA makes the same point in its guidance: hashing a MAC address without additional measures is pseudonymisation, not anonymisation; the hashing formula is in the controller's possession; adding a salt does not change that; storing multiple location points per device means the data are not anonymous. [DPA] Autoriteit Persoonsgegevens, "Wifi and bluetooth", Quick answers.

[Technical] The practical implication for Heilbronn is a staged analysis:

Stage What is processed Legal character Consequence
Passive radio reception 802.11/BLE frames; MAC in the frame header MAC is a device identifier; likely personal data when combined with location/time Reception is processing; a legal basis is needed if personal data
Raw MAC in sensor RAM Full MAC for milliseconds Personal data; transient Must be minimised; no logging, no transmission, no packet capture
Dedup identifier Hash or 16-bit ID; up to 1 hour Pseudonymous personal data while the key/raw data exist Legal basis; retention limit; no cross-sensor/cross-day linkage
Count per interval Integer Aggregate; may be anonymous if no small-cell inference Still assess; ePrivacy may apply to the collection; DPIA
Published CSV/Grafana Counts per sensor/15 min Potentially personal data in small cells; not publication-safe Coarsen, suppress, add noise before release

The table is the single most important analytical move in this note: it prevents the common error of arguing from the anonymity of the output to the lawfulness of the input.

Terminology: "frequency measurement" and "Bluetooth tracking." German practice uses "Passantenfrequenzmessung" or "Besucherstrommessung" for footfall measurement. That is a purpose, not a technology. The legal analysis depends on the sensor: a light barrier, a thermal/radar counter, or an on-device video counter that does not identify devices may avoid the GDPR/ePrivacy issues altogether; the CNIL expressly distinguishes simple footfall counters (a mat that counts steps, a turnstile) from devices that collect personal data. [Guidance] CNIL, 24 Feb 2020. "Bluetooth tracking" is likewise a family of techniques: passive BLE scanning of advertising addresses (the Heilbronn design), active BLE scanning, beacon-based proximity, or app-based beacon scanning. The common legal issue is whether a device identifier and location/time are processed; the specific radio technology does not change the analysis. [Guidance] EDPB Guidelines 2/2023, paras 57–60 (IoT/BLE examples); [DPA] AP, "Wifi and bluetooth".

2.4 MAC randomisation and BLE

[Technical] Modern devices randomise their Wi-Fi MAC addresses. Apple uses a private Wi-Fi address per SSID on iOS 14/iPadOS 14/macOS 14 and later; Android 10+ randomises by default for new networks; BLE uses resolvable private addresses that rotate, typically every 15 minutes. [Technical] Apple Platform Security, "Wi-Fi privacy"; Android Open Source Project, "MAC randomization"; Martin et al., PETS 2017; Becker et al., PETS 2019.

Randomisation reduces linkability but does not eliminate it:

  • some devices still use the hardware MAC in certain frames or configurations;
  • probe-request randomisation policies vary by OS, device model, and power state;
  • BLE advertisers may include identifying tokens in the payload or use static addresses; resolvable private addresses can be resolved by bonded devices, and address changes may be asynchronous with payload changes; [Technical] Becker et al., PETS 2019;
  • fingerprinting beyond the MAC (sequence numbers, information elements, RSSI patterns) can re-identify a device across randomisation cycles; [Technical] Vanhoef et al., AsiaCCS 2016; Martin et al., PETS 2017;
  • randomisation introduces measurement bias: the counted population is "devices currently using a locally administered address that the scanner hears," not all devices or people. [Technical] Code audit, §2.1.

Legal implication. MAC randomisation is a privacy feature outside the controller's control. It is not an anonymisation technique the controller can rely on. It may reduce the risk and the accuracy of the count, but the controller must still assess the data it actually processes, including the raw MAC before randomisation is observed and the hashed ID during the dedup window. [Guidance] EDPB Guidelines 02/2026, para. 25; [DPA] AP, "Wifi and bluetooth".


3. ePrivacy Directive and § 25 TDDDG

3.1 The EU rule

[Law] ePrivacy Directive 2002/58/EC, Art. 5(3), as amended by Directive 2009/136/EC, provides:

"Member States shall ensure that the storing of information, or the gaining of access to information already stored, in the terminal equipment of a subscriber or user is only allowed on condition that the subscriber or user concerned has given his or her consent, having been provided with clear and comprehensive information, in accordance with Directive 95/46/EC, inter alia, about the purposes of the processing. This shall not prevent any technical storage or access for the sole purpose of carrying out the transmission of a communication over an electronic communications network, or as strictly necessary in order for the provider of an information society service explicitly requested by the subscriber or user to provide the service."

[Law] Recital 24 explains that the terminal equipment of users and the information stored in it are part of the private sphere; hidden identifiers and other similar devices can enter terminal equipment without the user's knowledge and should be prohibited. Recital 25 refers to "cookies or similar devices." Recital 66 (added by Directive 2009/136/EC) says the user's consent should be required for storing information in terminal equipment or accessing information already stored. The proposed ePrivacy Regulation has not been adopted; the Directive remains the applicable EU instrument.

[CJEU] Planet49 confirms that Art. 5(3) applies regardless of whether the information is personal data, and that the consent standard is the GDPR standard: active behaviour, specific information, no pre-ticked boxes. [CJEU] C-673/17, paras 52–65. The EDPB reads the provision as technology-neutral, covering any information stored or accessed. [Guidance] EDPB Guidelines 2/2023, paras 9–12.

[Guidance] The EDPB's Guidelines 2/2023 are the most detailed technical interpretation. Key points:

  • "Information" includes non-personal data and personal data, regardless of who stored it. [Guidance] EDPB Guidelines 2/2023, paras 9–12.
  • The notion of "stored information" includes information stored by the user, by a hardware manufacturer (expressly "such as the MAC addresses of network interface controllers"), or by any other party. [Guidance] Ibid., para. 39.
  • "Gaining access" includes reading a value that the user did not store; the WP29 Opinion 9/2014 example is "requesting the MAC address of a network interface via the OS API." [Guidance] Ibid., para. 11, citing WP29 Opinion 9/2014, p. 8.
  • Routing identifiers such as the MAC or IP address of the terminal equipment, communicated to remote actors, can trigger Art. 5(3); the fact that the receiving entity is not the entity instructing the sending does not preclude application. [Guidance] Ibid., para. 42.
  • Local processing entirely inside the terminal equipment, with no information leaving the device, is outside Art. 5(3). [Guidance] Ibid., para. 44.
  • IP-only tracking can trigger Art. 5(3) where the IP address originates from the terminal equipment; the controller must assess. [Guidance] Ibid., paras 54–56.

3.2 The German rule

[Law] § 25 TDDDG (formerly § 25 TTDSG, in force since 1 December 2021; renamed on 14 May 2024) provides:

"(1) The storage of information in the terminal equipment of the end user or the access to information already stored in the terminal equipment shall only be permitted if the end user has consented on the basis of clear and comprehensive information. The information and consent shall be provided in accordance with Regulation (EU) 2016/679. (2) Consent under paragraph 1 is not required if (1) the sole purpose of the storage or access is the transmission of a communication over a public telecommunications network, or (2) the storage or access is strictly necessary for the provider of a digital service to provide a digital service explicitly requested by the user."

[Law] § 2(2) No. 6 TDDDG defines "terminal equipment" broadly: any equipment directly or indirectly connected to the interface of a public telecommunications network for sending, processing or receiving messages; the connection may be wire, optical fibre or electromagnetic. A smartphone with Wi-Fi or BLE enabled is terminal equipment; it is connectable to a public network (directly or through a router).

[Law] § 25 applies regardless of whether the information is personal data. [Guidance] DSK, OH Digitale Dienste (Nov 2024), para. 26 ("Im Unterschied zu den datenschutzrechtlichen Vorschriften begründet § 25 Abs. 1 TDDDG ein Einwilligungserfordernis ... unabhängig von einem Personenbezug"). The DSK guidance also says that, for mobile devices, "access to hardware device identifiers, advertising identifiers, phone numbers, SIM serial numbers (IMSI), contacts, call lists, Bluetooth beacons or SMS communication" must be considered, and "for all devices, the reading of the unique identifiers of the network hardware (MAC addresses) must be considered." [Guidance] DSK, OH Digitale Dienste, para. 22.

[Law] Enforcement: § 28(1) No. 13 TDDDG makes it an administrative offence to store or access information contrary to § 25(1); § 28(2) sets a fine of up to €300,000 for private offenders. [Law] § 28(4) provides that no fines are imposed on public authorities and other public bodies within the meaning of § 2(1) or (2) BDSG. [Law] § 29(2) allocates § 25 supervision over telecom providers and federal public bodies to the BfDI; for a municipal body in Baden-Württemberg, the competent authority is the LfDI Baden-Württemberg.

This is the decisive question for Heilbronn. The elements are:

Element ePrivacy Art. 5(3) § 25 TDDDG Application to passive MAC sniffing
Information Any information, not only personal data Any information; expressly no personal data required A MAC address is information; the EDPB treats NIC MAC addresses as stored information ([Guidance] EDPB Guidelines 2/2023, para. 39)
Terminal equipment Equipment of a subscriber/user connected to a public network § 2(2) No. 6 TDDDG; broad A smartphone with Wi-Fi/BLE enabled is terminal equipment
Storage Storing in the terminal equipment Same The sensor stores nothing in the phone: no
Access Gaining access to information already stored "Zugriff" on stored information The sensor receives a broadcast frame; it does not instruct the phone, retrieve from its memory, or use an OS API. Arguably no "access"
Consent Prior, informed, active; narrow exceptions Same; § 25(2) exceptions If access: consent required; legitimate interest is not a substitute; signage/opt-out is not consent
Personal data required? No No (DSK OH para. 26) Even an anonymous output does not escape § 25 if the collection is an access

The case for "no access." The ordinary meaning of "access" and the German legislative materials point to an active retrieval. The draft TTDSG explanatory memorandum describes the provision as protecting end users from third parties who "store or read out" information on their terminal equipment; it uses "Abruf" (retrieval) and "Auslesen" (readout). [Law] BT-Drs. 19/27441, p. 38. The DSK guidance's examples are active: access to hardware identifiers, JavaScript fingerprinting, Bluetooth beacons. [Guidance] DSK, OH Digitale Dienste, paras 22–24. The EDPB's own examples of "gaining access" are an OS API request and software-instructed transmission; even where the EDPB says the receiver need not be the entity instructing the transmission (para. 42), the device is still transmitting to a remote actor. In passive sniffing, the sensor is an unintended receiver of a broadcast, not an addressee of a communication. [Guidance] EDPB Guidelines 2/2023, paras 11, 42, 44. If every radio receiver "accessed" every device that broadcasts a MAC, ordinary Wi-Fi operation would require consent from every passer-by; that is not a sensible reading of Art. 5(3).

The case for "access." The EDPB's functional approach is broad: it says information includes MAC addresses of network interface controllers; routing identifiers such as the MAC address communicated to remote actors can trigger Art. 5(3); and the fact that the receiving entity is not the one instructing the sending does not preclude application. [Guidance] EDPB Guidelines 2/2023, paras 39, 42. A passive sensor is a receiving entity; the device sends the MAC; the sensor obtains the information. The purpose of Art. 5(3) is to protect the private sphere of terminal equipment against hidden tracking; passive MAC tracking is exactly the kind of invisible tracking the rule targets. The DSK guidance does not limit "Auslesen" to an OS API; it says "bei allen Geräten ist ... das Auslesen der eindeutigen Kennungen der Netzwerk-Hardware (MAC-Adressen) zu berücksichtigen." [Guidance] DSK, OH Digitale Dienste, para. 22. The Dutch DPA's guidance asks whether the sensor "gains access to information in a user's peripheral equipment because of the way the WiFi or Bluetooth sensors read data from them" and says such access requires explicit consent. [DPA] AP, "Legal bases for WiFi tracking and Bluetooth tracking". The AP's guidance is nuanced: it says access to information in peripheral equipment requires explicit consent, while also describing a legitimate-interest route for Wi-Fi/Bluetooth tracking. The tension reflects the interaction between the Dutch Telecommunications Act's Art. 11.7a (the national Art. 5(3) implementation) and the GDPR. It is another reason to obtain local advice rather than transplanting the Dutch model.

Assessment. There is no published EU or German decision on passive Wi-Fi/BLE sniffing under Art. 5(3)/§ 25. The better textual and purposive reading — in this author's assessment, not a settled ruling — is that passive reception is not "access" because the sensor does not interact with the terminal equipment's storage; the device broadcasts the frame for network discovery, not in response to the sensor. But the EDPB's broad language and the DSK's MAC example create a material risk that a supervisory authority or court would take the opposite view. The risk is higher if the sensor:

  • sends probe requests, association requests, or any active query;
  • uses an OS API or an app on the device to read the MAC;
  • captures payload or other stored information;
  • is presented as a "Wi-Fi hotspot" or otherwise induces the device to connect;
  • combines the MAC with other data to build profiles.

The Heilbronn design, as described, is at the favourable end: passive BLE scan, no scan requests, no association, hashed addresses kept on the sensor, only counts transmitted. [Technical] Code for Heilbronn blog; code audit §1.1. That makes the "no access" argument stronger, but it does not eliminate the uncertainty. A written § 25 memo should:

  1. describe the exact frames received and confirm that no active request is sent;
  2. confirm that no data are written to the device and no OS API is used;
  3. explain why receiving a broadcast frame is not "access to information already stored" (no retrieval from storage; the device initiates the transmission; the frame is not the stored information);
  4. address the EDPB's paras 39 and 42 and the DSK's para. 22, and explain why they do not cover passive reception;
  5. state the residual risk and the fallback if the authority disagrees.

If the memo cannot be made robust, the project should assume § 25 applies. In that case, consent is required. Consent at city scale is not realistically obtainable: the sensor collects from all passers-by, including those who have not installed an app, have not seen a sign, or have not taken any active step. [CJEU] Planet49, paras 52–65; [Law] GDPR Recital 32. A sign, an opt-out, or leaving Wi-Fi on is not consent. The only remaining routes are a statutory legal basis (for a public body) or a redesign that does not extract MAC addresses.

3.4 Notification and transparency

Short answer. If Art. 5(3)/§ 25 applies, prior notification is mandatory, but notification is not consent. If Art. 5(3)/§ 25 does not apply, notification is still required under GDPR Art. 14 (and Art. 21(4) for the right to object) whenever personal data are processed; if the data are truly anonymous, there is no GDPR notification duty, but transparency remains good practice and may be required by ePrivacy if consent is needed. The rest of this section explains the detail.

If Art. 5(3)/§ 25 applies, the controller must provide "clear and comprehensive information" before consent. A sign that merely says "Wi-Fi tracking" is not enough; it must identify the controller, the purposes, the data, the retention, the rights, and how to consent or refuse. But information is not consent; the consent must be active and prior.

If Art. 5(3)/§ 25 does not apply, the GDPR transparency regime still applies to personal data:

  • Art. 13 if the data are obtained from the data subject; Art. 14 if not. Passive sensing is arguably "not obtained from the data subject" in the ordinary sense, but the Conseil d'État characterized MAC collection as "direct collection" because the data are collected directly from the devices, even though the persons take no active step. [Enforcement] CE, 8 Feb 2017, no. 393714, para. 5. Under the GDPR, the safer approach is to provide Art. 13-style information at the point of collection, via on-site signage and a QR link to the full notice.
  • Art. 14(5)(b) allows an exception where providing information would involve a disproportionate effort. The exception is interpreted narrowly; for a fixed sensor network in public space, signage is feasible, so the exception is unlikely to apply. [Law] GDPR Art. 14(5)(b).
  • Art. 21(4) requires the right to object to be brought to the data subject's attention explicitly and separately, at the latest at the time of the first communication, where processing is based on Art. 6(1)(e) or (f). Signage should state the right to object and how to exercise it.

If the output is truly anonymous, the GDPR information duties do not apply to the output. But the collection stage still processes personal data (unless the sensor never extracts a MAC), so information is still needed for the collection. And if § 25 applies, the ePrivacy information duty applies regardless of anonymity.


4. GDPR obligations

4.1 Who is the controller?

[Law] Art. 4(7) defines the controller as the natural or legal person, public authority, agency or other body which alone or jointly with others determines the purposes and means of the processing. Art. 4(8) defines the processor as a person processing personal data on behalf of the controller. Art. 26 applies to joint controllers; Art. 28 to processors.

[CJEU] Joint controllership is broad. In Wirtschaftsakademie, the operator of a Facebook fan page was a joint controller with Facebook for the collection and processing of visitor data, even though it did not have access to the data, because it influenced the purposes and means. [CJEU] C-210/16, paras 28–39. In Fashion ID, a website operator that embedded a Facebook "Like" button was a joint controller for the collection and transmission of the data, but not for Facebook's subsequent processing. [CJEU] C-40/17, paras 65–71. The EDPB's Guidelines 07/2020 emphasize that joint controllership requires a common decision on purposes and means, or a convergence of decisions; it does not require equal responsibility or access. [Guidance] EDPB Guidelines 07/2020.

For Heilbronn, the roles are not determined by labels such as "citizen project," "open data," or "city partner." They depend on who decides:

  • why the data are collected (traffic planning, cleaning, urban development, research, commercial analytics);
  • which sensors are deployed, where, for how long, and at what granularity;
  • what is retained, published, and shared;
  • who can access the raw or pseudonymous data;
  • who benefits from the processing.

If the city determines the purposes and the Verein implements them under instruction, the city is controller and the Verein may be processor. If the Verein decides the purposes and means and the city merely receives the output, the Verein is controller (and needs its own legal basis). If they jointly decide, they are joint controllers and need an Art. 26 arrangement. If the Verein uses the data for its own purposes (e.g., open-data publication, research, community dashboards) in addition to the city's purposes, it is a controller for those purposes, even if it is a processor for the city's purposes. [Law] Art. 28(10).

[Guidance] The EDPB's Guidelines 07/2020 and the CJEU case law make clear that a written contract cannot disguise the reality: if the "processor" determines purposes or means, it is a controller. [Guidance] EDPB Guidelines 07/2020, paras 20–30; [Law] Art. 28(10).

4.2 Lawful basis

Public body. [Law] Art. 6(1)(f) does not apply to processing carried out by public authorities in the performance of their tasks. A municipality must rely on Art. 6(1)(e) (public task) read with Art. 6(3), which requires a legal basis in Union or Member State law, or on consent under Art. 6(1)(a). The legal basis must be clear, precise and foreseeable; a general competence to promote urban development or tourism may not be sufficient for continuous, covert MAC tracking. The city should identify the specific statutory task and document why the processing is necessary and proportionate. [Law] GDPR Art. 6(1)(e), 6(3); Recital 45. In Baden-Württemberg, the municipal framework is the Gemeindeordnung and the Landesdatenschutzgesetz (LDSG BW); the specific basis for footfall measurement should be confirmed by counsel.

Private association or company. [Law] Art. 6(1)(f) permits processing that is necessary for the purposes of a legitimate interest pursued by the controller or a third party, unless the data subject's interests or fundamental rights and freedoms override. The test is threefold: (1) a legitimate interest; (2) necessity and proportionality; (3) balancing. [CJEU] M5A, paras 37–44; [CJEU] KNLTB, paras 40–49; [CJEU] Meta, paras 42–47. A commercial interest can be legitimate, but it is not automatically sufficient; the processing must be necessary and the balancing must favour it. [CJEU] KNLTB, para. 49.

For public-space footfall tracking, the balancing is difficult:

  • people do not expect their phone's radio emissions to be collected by a third party;
  • they cannot practically avoid the measurement area;
  • the processing is invisible;
  • the data can reveal movement and presence;
  • less intrusive alternatives exist (manual counts, light barriers, thermal/radar sensors, on-device video counting);
  • the German and Dutch DPAs have repeatedly found the data subject's interests prevail. [DPA] ULD TB 2019, pp. 97–99; [DPA] AP, "Legal bases for WiFi tracking and Bluetooth tracking".

Safeguards — immediate anonymisation, no raw retention, per-window hashing, k-threshold, DP noise, spatial/temporal limits, opt-out — can improve the balancing, but they cannot cure a lack of necessity if a non-identifying sensor can meet the purpose. [CJEU] M5A (necessity/proportionality); [Guidance] EDPB Guidelines 01/2025, para. 23 (pseudonymisation can reduce risk and support legitimate interests, but does not create a legal basis).

Consent. [Law] Consent must be freely given, specific, informed and unambiguous, by a statement or clear affirmative action. [Law] Art. 4(11); Art. 7; Recital 32. Silence, pre-ticked boxes, inactivity, and "continuing to browse" are not consent. [CJEU] Planet49, paras 52–65. In a public-space passive sensing system, consent is practically unworkable: the controller cannot obtain prior consent from every passer-by, and an opt-in app covers only a self-selected subset. A sign is information, not consent. An opt-out is not consent. Leaving Wi-Fi or Bluetooth on is not consent. The Hamburg DPA warns that an enabled interface cannot be treated as consent to tracking. [DPA] HmbBfDI TB 2014/2015, p. 142.

Special categories. [Law] Art. 9 prohibits processing special categories of personal data unless an exception applies. Footfall counts are not inherently sensitive, but location data can reveal sensitive information when a sensor is placed near a hospital, clinic, place of worship, shelter, school, or political event. The DPIA should assess whether the sensor locations or the combination with other data could create Art. 9 data; if so, the processing needs an Art. 9(2) basis or must be redesigned. [Law] Art. 9; Recital 51.

4.3 Transparency and signage

[Law] Art. 12 requires information to be concise, transparent, intelligible and easily accessible, in clear and plain language. Art. 13/14 specify the content: identity and contact details of the controller and DPO; purposes; legal basis; categories of data; recipients; retention; rights; right to object; right to complain to a supervisory authority; whether provision is statutory/contractual; automated decision-making (not applicable).

For a public-space deployment, the practical standard is:

  • signs at every entrance to the measurement zone, before the zone, not inside it;
  • a short plain-language statement of purpose and data;
  • a QR code to the full privacy notice;
  • German and English (at least);
  • accessible formats (large print, plain language);
  • the identity of the controller and the DPO;
  • the legal basis (public task or legitimate interest);
  • the retention periods;
  • the right to object and how to exercise it;
  • the opt-out mechanism;
  • the complaint authority (LfDI Baden-Württemberg);
  • a statement of what is not collected (e.g., no names, no payload, no raw MAC retention).

[Guidance] The ICO's Wi-Fi location analytics guidance (2016, UK; no longer available at the original ICO URL) recommended a privacy impact assessment, notices at the point of collection and on the website, a defined retention period, and a simple opt-in/opt-out. The current ICO anonymisation and DPIA guidance applies the same principles. [Guidance] ICO, "Wi-Fi location analytics" (2016), as summarised in Bird & Bird (2016); ICO, "Anonymisation"; ICO, "Data protection impact assessments". The ICO is not binding in Germany, but its checklist is consistent with GDPR Art. 12–14 and Art. 35.

[Guidance] The CNIL recommends a two-level information model: a short first-level notice (purpose, controller, rights, opt-out) and a fuller second-level notice (all Art. 13/14 information). [Guidance] CNIL, 24 Feb 2020. This is a useful template for the Heilbronn signage.

Do not overclaim anonymity. A sign that says "fully anonymous" while the sensor processes raw MACs and keeps hashed IDs for an hour is misleading. The notice should distinguish the stages: "The sensor reads device addresses in memory, converts them into a temporary pseudonymous identifier, keeps that identifier for at most X minutes for deduplication, and transmits only counts. Raw addresses are not stored or transmitted." If the claim cannot be substantiated, do not make it.

4.4 Data minimisation and storage limitation

[Law] Art. 5(1)(c) requires personal data to be adequate, relevant and limited to what is necessary; Art. 5(1)(e) requires retention no longer than necessary. The controller must be able to justify each data element and each retention period.

For the Heilbronn pipeline:

Data Recommended handling Legal rationale
Raw MAC Memory only; deleted immediately after dedup (milliseconds); never logged, never transmitted, no packet capture Minimisation; reduces breach impact; supports the argument that the sensor does not "store" information
Hashed/pseudonymous ID Per-sensor, per-window keyed hash; retained only for the dedup window; key destroyed after the window; no cross-sensor or cross-day linkage Pseudonymous data; retention limitation; prevents movement profiles
Dedup window No longer than necessary for the count; if the output is 15-minute counts, a 15-minute window is the natural default; a 1-hour window needs a documented reason The current 1-hour cache is longer than the 15-minute reporting interval; it increases linkability
Counts Aggregate at the edge; transmit only counts; no per-device records Data minimisation; the output may be anonymous
Published data Zone × hour (or coarser); k-threshold; suppression; DP noise; no per-sensor minute data; no repeat-visit metrics Anonymisation/PETs; see §5
Logs No MACs; only health/telemetry (battery, uptime, error counts) Security without creating a new personal-data store

[Guidance] The CNIL's 2020 guidance provides a useful benchmark: if the data are anonymised within minutes, the CNIL recommends a maximum of five minutes before no identifying data are retained; if the data are pseudonymised at collection and anonymised or destroyed by the end of the day (in practice, within 24 hours), the CNIL accepts that regime; in all other cases, consent is required. [Guidance] CNIL, 24 Feb 2020. The CNIL also recommends not collecting precise timestamps for repeat-visit measurement, limiting the geographic area, and deleting pseudonymised data at the end of the measurement. [Guidance] Ibid. These are French recommendations, not German law, but they are a sensible benchmark for a DPIA.

[Guidance] The Hamburg DPA's 2014/2015 report says WLAN/Bluetooth tracking is highly intrusive and requires a legal basis, transparency before entering the area, and consent where applicable; it warns that an enabled interface is not consent. [DPA] HmbBfDI TB 2014/2015, pp. 141–143. It does not set a maximum retention period, but the logic supports short retention and no profiles.

[Guidance] The ULD's 2019 report is more specific: tracking based on passively captured probe requests for visitor counting in public spaces has no legal basis; a legitimate-interest balancing leads to the data subjects' interests prevailing because the data are collected without their knowledge and they have practically no way to prevent it. [DPA] ULD TB 2019, pp. 97–99. The ULD notes that simply disabling probe requests is difficult or impossible on some operating systems; often only airplane mode stops them. [DPA] Ibid., p. 98. This is directly relevant to the "opt-out by turning off Wi-Fi" argument: it is not a realistic opt-out.

4.5 Data protection impact assessment

[Law] Art. 35(1) requires a DPIA where a type of processing is likely to result in a high risk to the rights and freedoms of natural persons. Art. 35(3) lists three cases in particular: (a) a systematic and extensive evaluation of personal aspects relating to natural persons, based on automated processing including profiling, on which decisions with legal or similarly significant effects are based; (b) large-scale processing of special categories of data or criminal-offence data; and (c) systematic monitoring of a publicly accessible area on a large scale. The Heilbronn deployment falls under (c). [Law] GDPR Art. 35(3)(c).

Assuming the deployment processes personal data — which it does if it extracts MAC addresses for deduplication — it meets the criterion:

  • systematic: continuous, planned, automated sensing over months;
  • publicly accessible area: inner-city streets and squares;
  • large scale: a large number of passers-by over months, multiple sensors, and a substantial geographic extent; the exact number of data subjects is not in the delivered data, but Recital 91 lists the number of data subjects, the volume of data, the duration, and the geographic extent as relevant factors. [Law] Recital 91.

[Guidance] The Article 29 Working Party's DPIA guidelines (WP248 rev.01) list nine criteria; two or more indicate a high risk. The relevant ones here are: systematic monitoring; large-scale processing; innovative technology; matching or combining datasets; data concerning vulnerable data subjects; data not obtained from the data subject; invisible processing; and preventing data subjects from exercising a right or using a service. [Guidance] WP29, WP248 rev.01, pp. 10–12. A public-space MAC tracker hits at least five.

[Guidance] The LfDI Baden-Württemberg's private-sector "Muss-Liste" (2018) explicitly includes:

  • No. 2: "Extensive processing of data about the presence of persons" — example: "Offline tracking of customer movements in department stores, shopping centres or similar" — "a company processes GPS and WLAN data of passers-by and customers in order to track routes and purchasing behaviour." [Guidance] LfDI BW, Muss-Liste, No. 2.
  • No. 10: "Non-intended use of sensors of a mobile device in the possession of the data subject or of radio signals sent by such devices to determine the location or movement of persons over a substantial period" — same example. [Guidance] LfDI BW, Muss-Liste, No. 10.

This is directly on point: a private operator in Baden-Württemberg must conduct a DPIA for this processing. For a public body, there is no positive list for the public sector in Baden-Württemberg; the Ministry of Justice's 2024 threshold analysis states: "In Baden-Württemberg there is no positive list for the public sector." [Guidance] LfDI BW/MJM BW, Schwellenwertanalyse (29 Nov 2024). But Art. 35(3)(c) applies directly, and the risk assessment under Art. 35(1) may still require a DPIA. A public body should not treat the absence of a list as a negative list.

[Law] The DPIA must contain at least: a systematic description of the processing and purposes; an assessment of necessity and proportionality; an assessment of the risks; and the measures envisaged to address the risks. [Law] Art. 35(7). The controller must seek the DPO's advice (Art. 35(2)), and may seek the views of data subjects or their representatives (Art. 35(9)). If the residual risk remains high, the controller must consult the supervisory authority before processing (Art. 36). The DPIA must be reviewed when the processing changes (Art. 35(11)).

For Heilbronn, the DPIA should specifically cover:

  • the legal basis and the § 25 TDDDG analysis;
  • the exact data flow, including the raw MAC, hash length, key management, cache duration, and aggregation rule;
  • the re-identification risk, including small-cell counts, differencing, auxiliary data (camera, payment, hotspot, loyalty), and MAC randomisation;
  • the necessity and alternatives (manual counts, light barriers, thermal/radar, on-device video counting);
  • the transparency and opt-out mechanisms, and their effectiveness;
  • the roles of the city, the Verein, the LoRaWAN network operator, and any hosting provider;
  • the retention schedule and deletion verification;
  • the security measures and breach response;
  • the residual risk and whether Art. 36 prior consultation is needed;
  • a published non-technical summary, if possible.

[Guidance] The BfDI's DPIA page and the ICO's DPIA guidance provide checklists; the BfDI is not the competent authority for Heilbronn, but its methodology is the federal reference. [Guidance] BfDI, "Datenschutz-Folgenabschätzungen und Listen von Verarbeitungsvorgängen"; ICO, "Data protection impact assessments".

4.6 Security and breach

[Law] Art. 5(1)(f) requires integrity and confidentiality; Art. 32 requires technical and organisational measures appropriate to the risk, including pseudonymisation and encryption, resilience, availability, and regular testing. For an edge sensor network, the relevant measures include:

  • secure boot, signed firmware, and secure updates;
  • no default passwords; unique credentials; network segmentation;
  • encryption at rest and in transit (LoRaWAN payload encryption is not a substitute for end-to-end application-layer protection if the data are sensitive);
  • physical tamper protection;
  • no raw MAC storage; no packet capture; no debug logs containing identifiers;
  • access control and audit logging for the backend;
  • a vulnerability disclosure and incident response process;
  • vendor security attestations (ISO 27001, SOC 2 Type II, or equivalent);
  • regular penetration testing and firmware review.

[Law] A personal data breach must be notified to the supervisory authority within 72 hours where there is a risk to rights and freedoms (Art. 33), and to data subjects where the risk is high (Art. 34). A processor must notify the controller without undue delay (Art. 33(2)). If the data are truly anonymous, the GDPR breach regime does not apply to the anonymous output; but the collection stage may still involve personal data, and a breach of the sensor or backend could expose pseudonymous identifiers. The incident response plan should cover both.

4.7 Data subject rights and opt-out

[Law] Data subjects have rights of access, rectification, erasure, restriction, objection, and portability (Art. 15–20), plus the right not to be subject to automated decision-making (Art. 22, not applicable here) and the right to complain to a supervisory authority (Art. 77). Art. 11 provides a limited exception: if the controller can demonstrate that it cannot identify the data subject, it may not be obliged to comply with Arts. 15–20, but it must inform the data subject if possible and must assist if the data subject provides additional information enabling identification.

For a MAC tracking system, the controller may not be able to link a person to a MAC. The practical rights are therefore:

  • Objection (Art. 21): if the legal basis is Art. 6(1)(e) or (f), the data subject can object; the controller must stop unless it demonstrates compelling legitimate grounds. The right must be brought to the data subject's attention explicitly and separately at the latest at first communication (Art. 21(4)). Signage should state it.
  • Opt-out: a mechanism to stop counting a device. The CNIL recommends a dedicated Wi-Fi network named wifi_tracking_optout or a website/app where the user enters the MAC address; turning off Wi-Fi is not a satisfactory opt-out. [Guidance] CNIL, 24 Feb 2020. For BLE, a MAC-based opt-out is technically fragile because of randomisation; a dedicated opt-out SSID only works for Wi-Fi. A practical BLE opt-out may be a web form plus a commitment to delete existing pseudonymous data and suppress future matches; but the controller should be honest about its limits.
  • Access/erasure: if the controller cannot identify the data subject, Art. 11 may limit the response; but the controller should have a process and explain it in the notice.
  • Complaint: the LfDI Baden-Württemberg; the notice should give the address.

[Guidance] The CNIL's position is that for short-delay anonymisation, the access, rectification and restriction rights are not useful because the controller does not need to identify the persons; the right to object must still be provided. [Guidance] CNIL, 24 Feb 2020. This is a sensible model for a privacy-preserving design, but it does not remove the need for a legal basis or the Art. 21 notice.

4.8 Processors, joint controllers, and transfers

[Law] Art. 28 requires a written contract with a processor, including: processing only on documented instructions; confidentiality; security; sub-processor authorisation; assistance with data subject rights and DPIAs; breach notification under Art. 33(2); deletion or return at the end; audit and inspection. Art. 28(4) applies the same obligations to sub-processors. Art. 28(10) provides that a processor that determines purposes and means becomes a controller for that processing.

[Law] Art. 26 requires joint controllers to arrange their responsibilities in a transparent manner, in particular for the exercise of data subject rights and the provision of information; the essence of the arrangement must be made available to data subjects. The arrangement does not affect the fact that data subjects can exercise their rights against either controller.

For Heilbronn, the contracts should cover:

  • the city/Verein/vendor roles and the purposes;
  • the exact data fields, retention, and aggregation rules;
  • a prohibition on secondary use, re-identification, profiling, model training, or sale;
  • a prohibition on transferring raw or pseudonymous data outside the EU/EEA without a legal transfer mechanism;
  • security requirements, audit rights, and incident notification;
  • deletion and return at the end, with certification;
  • sub-processor approval and flow-down obligations;
  • cooperation with the DPIA and prior consultation;
  • liability and indemnity.

[Law] International transfers require Art. 44–49. If a vendor or cloud provider is outside the EEA, the controller needs an adequacy decision (e.g., the EU–US Data Privacy Framework, if the recipient is certified), standard contractual clauses plus a transfer impact assessment, or another Art. 46 mechanism. [CJEU] Schrems II, C-311/18. For a local pilot, EU hosting and EU sub-processors are the simplest route.


5. Anonymisation and privacy-enhancing technologies

[Law] Recital 26 is the starting point: anonymous information is information that does not relate to an identified or identifiable person, or personal data rendered anonymous so that the data subject is not or no longer identifiable. The assessment considers all means reasonably likely to be used by the controller or another person, including costs, time, available technology, and technological developments.

[CJEU] Breyer makes the test relative and contextual: identification through third-party data can make data personal even if the controller alone cannot identify; only a disproportionate effort excludes identifiability. [CJEU] C-582/14, paras 42–46. EDPS v SRB adds that the same pseudonymised data can be personal for the sender and non-personal for a recipient who lacks the key, but the sender's own transparency and legal-basis obligations remain. [CJEU] C-413/23 P, paras 82–85.

[Guidance] The EDPB's 2026 draft Anonymisation Guidelines set out the current framework:

  • Anonymity is assessed from the perspective of each relevant entity, typically the party for whom the data are intended to be anonymous. [Guidance] EDPB Guidelines 02/2026, executive summary.
  • Data can relate to a person by content, purpose or effect; the link need not be readily apparent; aggregate data can still relate to individuals if techniques can reveal information about a single individual. [Guidance] Ibid., paras 17–19.
  • A hardware identifier such as a MAC address can identify a device and, in turn, the individual using it. [Guidance] Ibid., para. 25.
  • The three criteria are No Record Isolation, No Linkage, and No Inference. [Guidance] Ibid., §3.4.
  • Aggregate data will usually satisfy No Record Isolation if the aggregation was done correctly, but they must still pass No Linkage and No Inference. [Guidance] Ibid., para. 58.
  • Inference can produce record-level data from aggregate data, including through differencing, outlier analysis, and combining multiple aggregate datasets. [Guidance] Ibid., paras 76–78. The EDPB gives the example of salary aggregates that reveal an individual's salary by subtraction. [Guidance] Ibid., Example 16.
  • Re-identification techniques are less likely to be effective against aggregate data, but this depends on the amount of data aggregated and the auxiliary information available. [Guidance] Ibid., para. 86.
  • The guidelines do not approve any specific technique as a safe harbour; the assessment must be documented and reassessed as technology evolves. [Guidance] Ibid., paras 84–87.

[Guidance] The Article 29 Working Party's Opinion 05/2014 remains a useful taxonomy: anonymisation techniques include randomisation (noise, permutation, differential privacy) and generalisation (k-anonymity, l-diversity, t-closeness), plus suppression; the three risks are singling out, linkability, and inference. [Guidance] WP29 Opinion 05/2014, pp. 5–8. The Opinion also makes clear that hashing is pseudonymisation, not anonymisation, where the original values can be recovered or brute-forced. [Guidance] Ibid., pp. 20–21.

[Guidance] The ICO's current anonymisation guidance uses a similar risk-based approach, including a "motivated intruder" test, and distinguishes anonymisation from pseudonymisation. [Guidance] ICO, "Anonymisation"; ICO, "Pseudonymisation". The ICO is UK guidance, not binding in the EU, but it is consistent with the EDPB framework.

5.2 The operator's claim, examined

Code for Heilbronn's blog says: "These MAC addresses are hashed and stored for at most one hour. This means that no conclusions can be drawn about specific devices or persons and no profiles can be created. Moreover, these MAC addresses do not leave the sensors even hashed, because only the number of counted addresses is sent." [Technical] Code for Heilbronn, 18 Sep 2025 (author's translation).

The first sentence is a reasonable description of the design intent. The second sentence is too strong as a legal and technical claim:

  • Hashing is pseudonymisation, not anonymisation. The controller holds the hash function/key; the raw MAC is observable before hashing; the hashed value is a unique identifier during the retention window. [Guidance] EDPB Guidelines 01/2025, para. 22; [DPA] AP, "Wifi and bluetooth"; [Enforcement] CE, 8 Feb 2017, no. 393714, para. 8.
  • The hash space matters. A full SHA-256 hash of a 48-bit MAC is not a strong anonymisation technique because the input space is small and enumerable; an unsalted hash can be brute-forced. A 16-bit ID (the upstream reference) has only 65,536 values; collisions are frequent, and the ID is not a cryptographic hash. A keyed hash with a secret, rotated key is better, but it is still pseudonymous while the key exists. [Technical] Code audit, §§1.2, 2.2; [Guidance] WP29 Opinion 05/2014, pp. 20–21.
  • The one-hour cache creates a one-hour linkability window. If the sensor stores the hashed ID for an hour, it can link multiple detections of the same device within that hour. Whether that creates a "profile" depends on the data structure and the sensor's coverage; a bitmap of IDs per sensor is less revealing than a timestamped trajectory, but it is still linkability. The operator's claim of "no profiles" should be replaced with a precise description of what is retained and why.
  • The output counts can still be personal data in small cells. A 15-minute count of 1 at a sensor reveals that one device was present. If the sensor covers a small area or a sensitive location, that can be identifying. The count is not automatically anonymous because it is a number. [Guidance] EDPB Guidelines 02/2026, paras 17–19, 76–78.

Recommendation. Replace the absolute claim with a precise, verifiable statement:

"The sensor reads BLE advertising addresses in memory. It converts each address into a temporary keyed pseudonymous identifier, retains that identifier only for the deduplication window (maximum X minutes), and transmits only the number of distinct identifiers per interval. Raw addresses are never stored or transmitted. The published counts are aggregated to zones and time bins and are suppressed below a minimum count. This reduces identifiability; it is not a claim that the data are anonymous in every context."

This is defensible, transparent, and consistent with the legal standard.

5.3 k-anonymity and cell suppression

[Technical] k-anonymity, introduced by Sweeney, requires that each record be indistinguishable from at least k−1 other records with respect to the quasi-identifiers. [Technical] Sweeney, 2002. For footfall counts, there are two different uses:

  1. Record-level k-anonymity on the underlying device-location records. Generalise location (e.g., sensor zone instead of exact coordinates) and time (e.g., hour instead of minute), and suppress combinations with fewer than k devices. This reduces singling out but does not prevent linkage or inference.
  2. Cell suppression on aggregate counts. Publish a count only if it is at least k; suppress or merge cells below k; use complementary suppression so that suppressed cells cannot be recovered by subtraction from totals. This is the practical analogue for a count product.

[Guidance] There is no legal k threshold. The CNIL's 2020 guidance speaks of a "taux de collision élevé" (high collision rate) — many MAC addresses must map to the same identifier after anonymisation — rather than a fixed k. [Guidance] CNIL, 24 Feb 2020. The EDPB's 2026 draft does not set a k; it requires the three criteria. [Guidance] EDPB Guidelines 02/2026, §3.4. The ICO's guidance discusses k-anonymity as a technique with limitations, not a safe harbour. [Guidance] ICO, "Anonymisation".

For a Heilbronn-style product, a defensible starting point is:

  • publish zone × hour (or zone × day), not sensor × 15 minutes;
  • require k ≥ 10 (or 20 for sensitive locations/times) distinct devices per published cell;
  • suppress cells below k; merge them into a larger area/time bin or mark them as "below threshold";
  • apply complementary suppression so that a suppressed cell cannot be derived from published totals;
  • avoid overlapping zones that allow differencing;
  • avoid publishing repeat-visit rates or trajectories; these require linking devices across time and are incompatible with the anonymity objective;
  • document the choice of k and the suppression rules in the DPIA and in a public methodology note.

[Guidance] The EDPB's 2026 draft expressly warns that aggregate data can fail No Inference through differencing and combination; a k-threshold alone is not enough. [Guidance] EDPB Guidelines 02/2026, paras 76–78. The CNIL's requirement to avoid precise timestamps for repeat-visit measurement points in the same direction. [Guidance] CNIL, 24 Feb 2020.

5.4 Differential privacy

[Technical] Differential privacy (DP) provides a formal guarantee that the output of a computation is approximately the same whether or not any single individual's data are included. The standard mechanism adds calibrated noise (Laplace or Gaussian) to a query result; the privacy loss is controlled by ε (epsilon), with smaller ε meaning stronger privacy and more noise. [Technical] Dwork & Roth, 2014.

For footfall counts:

  • Sensitivity: for a count query, adding or removing one person changes the count by 1, so the sensitivity is 1.
  • Mechanism: Laplace noise with scale 1/ε, or a Gaussian mechanism with the appropriate σ.
  • Budget: every release consumes privacy budget; repeated queries compose. A dashboard that lets users query arbitrary time/space cells can exhaust the budget and leak information through differencing. Use a fixed set of published tables, a pre-defined query set, or a privacy budget tracker.
  • Local vs central: central DP at the edge/backend is more accurate; local DP would require adding noise on each device, which is not possible for a passive sensor that does not run code on the phone. Secure aggregation can avoid a trusted curator, but it is more complex.
  • Post-processing: clamping negative noisy counts to zero and rounding are allowed post-processing, but they can affect the guarantee; document the pipeline.
  • Utility: for small counts, DP noise can swamp the signal; combine with coarser aggregation and suppression. For large city-scale counts, the relative error is small.

[Guidance] DP is a PET, not a legal safe harbour. The EDPB's 2026 draft does not endorse a specific ε or declare DP outputs anonymous; it requires the No Isolation/No Linkage/No Inference assessment. [Guidance] EDPB Guidelines 02/2026, §3.4. The ICO's PETs guidance describes differential privacy as a technique that adds noise to produce aggregate statistics, while noting the need to assess the privacy budget and the utility trade-off. [Guidance] ICO, "Privacy-enhancing technologies (PETs)". A DP output with a large ε, or with repeated queries, can still be personal data.

Recommendation. Use DP as an additional layer, not the only layer. Publish only a fixed set of coarse tables; set a privacy budget per period; document ε and the composition; combine with k-threshold and suppression; do not expose an ad-hoc query interface over the raw counts. A reasonable starting point for a public dashboard is ε ≈ 0.1–1 per release, but the choice must be justified by the DPIA and the expected number of releases. This is an engineering recommendation, not a legal requirement.

5.5 MAC-specific PETs and key management

The following measures are the technical core of a defensible design:

  1. Passive only. Do not send probe requests, association requests, or any active query. Do not impersonate a hotspot. Do not use an OS API or an app to read identifiers. [Technical] Code audit §1.1 (passive BLE scan).
  2. No raw retention. Keep the raw MAC in volatile memory only for the minimum time needed to compute the dedup identifier; never write it to flash, logs, packet captures, or the LoRaWAN payload. [Technical] Code audit §3 (upstream design); Code for Heilbronn blog (only counts transmitted).
  3. Keyed hash, not a plain hash. Use a keyed MAC (HMAC-SHA-256) or a keyed pseudonymisation function with a secret key stored in a secure element. Do not use an unsalted hash; do not use a static global salt; do not truncate to 16 bits. [Guidance] ICO, "Pseudonymisation" (long salts/peppers; test computational feasibility); [Guidance] WP29 Opinion 05/2014, pp. 20–21.
  4. Per-sensor, per-window keys. Derive a different key for each sensor and each counting window (e.g., 15 minutes or 1 hour). Destroy the key at the end of the window. This prevents cross-sensor and cross-window linkage. [DPA] AP, "Legal bases for WiFi tracking and Bluetooth tracking" (recommends hashing differently for each measuring location); [Guidance] CNIL, 24 Feb 2020 (salt renewed regularly).
  5. No cross-sensor linkage. Do not send the same pseudonymous ID to multiple sensors or to the backend. Aggregate at the sensor; if the backend needs to combine counts, combine only counts, not identifiers.
  6. No trajectories. Do not store a sequence of sensor IDs per device; do not compute paths, dwell times, or repeat-visit rates. The JCDecaux case shows that a purpose requiring routes/repeat visits is incompatible with anonymisation. [Enforcement] CE, 8 Feb 2017, no. 393714, para. 8.
  7. Short dedup window. Align the window with the reporting interval; if the output is 15-minute counts, a 15-minute window is the natural default. A 1-hour window needs a documented justification. [Law] Art. 5(1)(e); [Guidance] CNIL, 24 Feb 2020.
  8. Edge aggregation. Transmit only counts. Do not transmit hashed IDs, even encrypted. This reduces the attack surface and supports the argument that the identifiers do not leave the sensor. [Technical] Code for Heilbronn blog; code audit §3.
  9. No fingerprinting. Do not use sequence numbers, information elements, RSSI patterns, or payload tokens to re-identify devices across MAC randomisation. [Technical] Vanhoef et al., 2016; Martin et al., 2017; Becker et al., 2019.
  10. Secure key management. Store keys in a secure element or trusted execution environment; restrict access; rotate; audit; zeroise on tamper. If the key is compromised, the pseudonymous data become linkable to raw MACs. [Law] Art. 32.

5.6 When do aggregated counts become anonymous?

There is no single threshold. The legal test is contextual: the counts are anonymous if they do not relate to an identified or identifiable person, considering all means reasonably likely to be used by the controller or another person. [Law] Recital 26; [CJEU] Breyer, paras 42–46. The following questions are a practical test:

  1. Is there any record-level data left? If the controller retains raw MACs, hashed IDs, per-device timestamps, or trajectories, the data are personal (or pseudonymous) regardless of what is published.
  2. Can a published cell be isolated to one person? If a cell has a count of 1, or a unique combination of attributes, it can single out an individual. Apply k-threshold and suppression.
  3. Can two published tables be linked to the same individual? If the controller publishes per-sensor and per-zone tables, or overlapping time bins, differencing can reveal a small cell. Use consistent geography, complementary suppression, and a fixed publication set.
  4. Can auxiliary data identify the person? Camera footage, payment records, hotspot logins, loyalty data, event attendance, or social media can be combined with a count to identify a person. The EDPB's No Linkage criterion requires assessing these means. [Guidance] EDPB Guidelines 02/2026, §3.4.2.
  5. Can the purpose or effect of the data relate to an individual? Even a count can relate to a person if it is used to treat them differently (e.g., to deny entry, target advertising, or infer sensitive behaviour). [Guidance] EDPB Guidelines 02/2026, paras 17–19.
  6. Is the anonymisation irreversible and documented? If the key is destroyed, the raw data are gone, and no linkage is reasonably possible, the data may be anonymous. Document the assessment, the threat model, and the reassessment schedule. [Guidance] EDPB Guidelines 02/2026, paras 84–87; [Guidance] ICO, "Anonymisation".

For the Heilbronn data, the current 15-minute per-sensor CSV is best treated as internal aggregate data that may still be personal data in small cells, not as a publication-safe anonymous product. Before public release, coarsen to zone × hour/day, apply k-threshold and suppression, consider DP noise, remove repeat-visit/trajectory metrics, and document the assessment. [Guidance] EDPB Guidelines 02/2026, paras 76–78; [Guidance] CNIL, 24 Feb 2020.

5.7 Suggested publication rules

Rule Suggested setting Why
Spatial resolution Zone (not individual sensor) Reduces singling out and differencing
Temporal resolution Hour or day (not 15 minutes) Reduces presence inference and differencing
Minimum cell count k ≥ 10; k ≥ 20 for sensitive locations/times Suppresses small cells; no legal k, but a defensible starting point
Complementary suppression Suppress additional cells so totals do not reveal suppressed cells Prevents subtraction attacks
DP noise Laplace/Gaussian; fixed query set; privacy budget Formal protection against inference; not a standalone legal basis
Repeat visits Do not publish Requires cross-time linkage; incompatible with anonymity
Trajectories Do not publish Requires cross-sensor linkage; JCDecaux shows the problem
Sensitive locations Exclude or coarsen Art. 9 risk; chilling effects
Methodology Publish a note Accountability; allows scrutiny
Reassessment Annual or on change Technology and auxiliary data evolve

These are engineering and disclosure-control recommendations, not legal advice. The DPIA should justify the chosen parameters.


6. Enforcement and regulatory practice

6.1 Cases and decisions

Jurisdiction Matter Facts Outcome Lesson
Netherlands Enschede Wi-Fi tracking [Enforcement] Municipality used at least 10 sensors in the city centre (25 May 2018–30 Apr 2020) to capture MAC addresses of Wi-Fi-enabled devices, pseudonymised them, stored a 6-month long-term table, and estimated unique visitors. AP fined €600,000 (decision 11 Mar 2021, announced 29 Apr 2021) for processing personal data without a legal basis. Rechtbank Overijssel annulled the fine on 2 Feb 2024 because the AP had not proved that the municipality processed personal data; the Raad van State upheld on 29 Jul 2026, essentially because the AP's new "direct identification" argument had not been in the original decision. The fine is off. The AP's theory (pseudonymised MAC + location = personal data) was not accepted on the evidence. The case is fact-specific and procedural; it is not a ruling that Wi-Fi tracking is lawful or anonymous. The regulator must prove identifiability; the controller should document its own assessment.
France JCDecaux / La Défense [Enforcement] Six Wi-Fi counting boxes on public advertising furniture, capturing MAC addresses within 25 m and calculating positions, for a four-week test. JCDecaux proposed truncating the last half-octet, salting, and keyed hashing. The CNIL refused authorisation (deliberation 2015-255, 16 Jul 2015). Conseil d'État, 8 Feb 2017, no. 393714, upheld the refusal. The court held that hashing/salting left the controller able to identify and link records, and that the purpose (routes and repeat passages) was incompatible with anonymisation. Pseudonymisation is not anonymisation; the purpose matters; a real opt-out and information are required. This is the strongest EU judicial warning against a "we hash it" defence.
Germany (Schleswig-Holstein) ULD offline tracking [DPA] Two city-marketing operators (a Verein and a GmbH) passively captured probe requests to measure visitor flows in two cities. The ULD opened administrative proceedings. The ULD's legal view: MAC addresses are personal data; there is no legal basis; legitimate interest fails because the data are collected without knowledge and people cannot practically prevent it; no consent. Both operators stopped; the proceedings were closed. This is the closest German analogue to a citizen-association footfall project. A Verein is not exempt. Passive probe-request tracking can be treated as unlawful even without a fine.
Germany (Bavaria) BayLDA offline-tracking review [DPA] 2016 review of 11 retailers; all denied using offline tracking. A complaint led to an on-site inspection of a fashion chain's Wi-Fi tracking test at a store entrance, where a sign had been posted. The company stopped the test after discussion; the BayLDA could not complete a technical/legal assessment. Supervisory attention is real; a sign at the entrance did not make the test lawful. The case shows the value of a prior DPIA and a legal basis.
Germany (Hamburg) HmbBfDI indoor tracking guidance [DPA] The 2014/2015 activity report discusses WLAN/Bluetooth indoor tracking of customers and passers-by. The DPA says such tracking massively interferes with rights and is only permitted under very narrow conditions; MAC addresses and movement profiles are personal data; a legal basis is required; transparency before entering the area is required; an enabled interface is not consent. A German DPA views indoor/outdoor WLAN/Bluetooth tracking as high-risk. The report predates the GDPR but its reasoning is consistent with the current framework.
Germany (federal) TDDDG fines [Law] § 28(1) No. 13 TDDDG makes a violation of § 25(1) an administrative offence; § 28(2) sets a fine of up to €300,000 for private offenders. § 28(4) excludes fines against public bodies. A private operator (including a Verein, depending on its status) faces a TDDDG fine; a public body faces supervisory measures, complaints, and reputational risk, not a TDDDG fine.

No published German fine specifically for passive Wi-Fi/BLE footfall tracking was found. The German cases were administrative/stop-and-desist. That is not evidence of legality; it may reflect the novelty of the technology, the cooperation of the operators, and the difficulty of proving personal data (as in Enschede).

Temporal caveat. The JCDecaux decision (2017) applies the French law implementing Directive 95/46/EC; the ULD (2019), BayLDA (2016) and Hamburg (2014/2015) materials predate the TTDSG/TDDDG and, in some cases, the GDPR. They are cited for their reasoning on MAC addresses, identifiability, legitimate interest and transparency, not as direct applications of the current German statute. The current framework is the GDPR plus § 25 TDDDG; the older decisions show how DPAs approach the technology, but they do not resolve the § 25 passive-sniffing question.

6.2 Regulatory guidance

  • CNIL (France), 24 Feb 2020. Three regimes for audience/footfall measurement in publicly accessible spaces: (1) anonymisation within minutes (CNIL recommends a maximum of five minutes) — legitimate interest, limited information, opt-out; (2) pseudonymisation at collection and anonymisation/destruction by the end of the day (max 24 hours) — legitimate interest with safeguards; (3) all other cases — consent. [Guidance] CNIL, 24 Feb 2020. The guidance also recommends deleting raw MACs immediately, hashing with a regularly renewed salt, deleting pseudonymised data at the end, limiting the area and duration, avoiding precise timestamps for repeat-visit measurement, and providing a real opt-out (dedicated SSID or website/app; turning off Wi-Fi is not satisfactory).
  • Autoriteit Persoonsgegevens (Netherlands). Wi-Fi/Bluetooth tracking data are personal data; hashing alone is pseudonymisation; storing multiple location points per device means the data are not anonymous; municipalities need an independent legal basis, specific periods, and precisely described areas; private parties may rely on legitimate interest only in limited cases (property protection, safety, crowd control), with proportionality, subsidiarity, immediate anonymisation, different hashing per location, spatial/temporal limits, and opt-out. [DPA] AP, "Wifi and bluetooth"; "Legal bases for WiFi tracking and Bluetooth tracking".
  • DSK (Germany), Nov 2024. § 25 TDDDG applies regardless of personal data; reading MAC addresses and Bluetooth beacons must be considered; the guidance follows EDPB Guidelines 2/2023. [Guidance] DSK, OH Digitale Dienste, paras 22, 26.
  • LfDI Baden-Württemberg. The private-sector Muss-Liste includes offline tracking of customer movements and the non-intended use of device sensors/radio signals to determine location or movement over a substantial period; a DPIA is mandatory for private controllers. There is no public-sector positive list in Baden-Württemberg. [Guidance] LfDI BW, Muss-Liste, Nos. 2, 10; LfDI BW/MJM BW, Schwellenwertanalyse (2024).
  • ICO (UK), 2016/current. The 2016 Wi-Fi location analytics guidance (no longer available at the original ICO URL) recommended a privacy impact assessment, notices at the point of collection, a defined retention period, and a simple opt-in/opt-out. The current anonymisation, pseudonymisation, PETs, and DPIA guidance applies the same principles. [Guidance] ICO, "Wi-Fi location analytics" (2016; summarised in Bird & Bird, 2016); ICO, "Anonymisation"; "Pseudonymisation"; "Privacy-enhancing technologies"; "Data protection impact assessments".

6.3 What the enforcement record means for Heilbronn

  1. The risk is real but the legal question is unsettled. Enschede shows a €600,000 fine for the exact use case; it also shows the fine can be annulled if the regulator cannot prove personal data. Do not treat either outcome as a safe harbour.
  2. German DPAs are skeptical of passive public-space MAC tracking. The ULD's 2019 position is the closest to Heilbronn: passively captured probe requests, visitor counting, a Verein and a GmbH, no legal basis, legitimate interest fails. The operators stopped.
  3. A sign is not a legal basis. The BayLDA inspected a store with a sign; the company stopped. The Hamburg DPA says an enabled interface is not consent. Signage is necessary but not sufficient.
  4. Hashing is not a magic defence. The French Conseil d'État rejected truncation + salting + keyed hashing because the controller could still identify/link and the purpose required it. The Dutch AP says hashing is pseudonymisation. The German DSK says § 25 applies regardless of personal data.
  5. A public body's main risk is not a TDDDG fine. § 28(4) TDDDG excludes fines against public bodies; but the city can face supervisory orders, Art. 58 GDPR measures, complaints, political scrutiny, and reputational damage. A private Verein or vendor can face a TDDDG fine up to €300,000 and GDPR fines.

7. Compliance checklist for a privacy-preserving architecture

This checklist is written for the Heilbronn/Paxcounter design. It is a practical starting point, not a legal opinion. Each item should be evidenced in the DPIA or the processing record.

  • Identify the controller(s). City, Verein, vendor, LoRaWAN network operator, hosting provider. Document who decides purposes and means. If joint, draft an Art. 26 arrangement; if processor, draft an Art. 28 DPA. [Law] Arts. 4(7), 26, 28.
  • Identify the legal basis. Public body: Art. 6(1)(e) + Art. 6(3) + the specific legal basis in BW law; Art. 6(1)(f) is unavailable. Private: Art. 6(1)(f) with a documented legitimate-interest assessment; consent only if genuinely obtainable. [Law] Art. 6.
  • Write the § 25 TDDDG memo. Describe the exact frames, confirm passive-only reception, confirm no OS API/active request, address EDPB Guidelines 2/2023 paras 39/42 and DSK OH para. 22, state the residual risk and the fallback. If the memo is not robust, assume § 25 applies and do not deploy without consent or a statutory basis. [Law] § 25 TDDDG; [Guidance] EDPB Guidelines 2/2023; DSK OH.
  • Check whether the purpose is a public task. If the city uses the data for cleaning, maintenance, urban development, or tourism, identify the statutory task and the necessity/proportionality analysis. [Law] Art. 6(1)(e), 6(3).
  • Do not use "open data" as a legal basis. Publishing counts is a separate purpose; assess compatibility (Art. 6(4)) or rely on a separate basis. If the output is anonymous, the GDPR does not apply to the output, but the collection still needs a basis.

B. DPIA and governance

  • DPIA before deployment. Art. 35(3)(c); LfDI BW Muss-Liste Nos. 2/10 for private operators; no public-sector list in BW, but Art. 35(3)(c) applies directly. [Law] Art. 35; [Guidance] WP248; LfDI BW.
  • Involve the DPO. Art. 35(2); document the advice. [Law] Art. 35(2), 39.
  • Assess necessity and alternatives. Compare with manual counts, light barriers, thermal/radar, on-device video counting, and sample-based surveys. Explain why MAC/BLE is necessary and why a less intrusive method is not sufficient. [CJEU] M5A; [Law] Art. 5(1)(c).
  • Assess re-identification risk. Small cells, differencing, auxiliary data, MAC randomisation, hash length, cache duration, cross-sensor linkage. Document the threat model and the residual risk. [Guidance] EDPB 02/2026; [Guidance] WP29 Opinion 05/2014.
  • Prior consultation if residual high risk. Art. 36; allow time for the LfDI BW to respond. [Law] Art. 36.
  • Record of processing. Art. 30; include the purpose, data categories, recipients, retention, security measures, and the § 25 analysis. [Law] Art. 30.
  • Review and audit. Art. 35(11); annual review or on change; independent audit of the sensor firmware and backend.

C. Architecture and PETs

  • Prefer a non-identifying sensor. If a light barrier, thermal sensor, radar, or on-device video counter can meet the accuracy requirement, use it. A MAC-free design avoids the core legal problem. [Guidance] CNIL, 24 Feb 2020 (distinguishes simple footfall counters that do not identify).
  • Passive only. No probe requests, no association requests, no hotspot impersonation, no OS API. [Technical] Code audit §1.1.
  • No raw MAC retention. Memory only; delete immediately after dedup; no logs, no pcap, no flash, no LoRaWAN payload. [Technical] Code for Heilbronn blog; code audit §3.
  • Keyed hash, per sensor, per window. HMAC-SHA-256 or equivalent; secret key in a secure element; rotate every window; destroy the old key. Do not use a plain hash, a static salt, or a 16-bit ID. [Guidance] ICO, "Pseudonymisation"; [DPA] AP, "Legal bases".
  • No cross-sensor/cross-window linkage. Aggregate at the sensor; transmit counts only; do not send the same pseudonymous ID to the backend. [DPA] AP, "Legal bases".
  • No trajectories or repeat-visit metrics. Do not store per-device paths, dwell times, or repeat visits. [Enforcement] CE, 8 Feb 2017, no. 393714, para. 8.
  • Short dedup window. Align with the reporting interval; document why a longer window is necessary. [Law] Art. 5(1)(e).
  • Edge aggregation. Counts only; no identifiers in the uplink. [Technical] Code for Heilbronn blog.
  • No fingerprinting. No sequence numbers, information elements, RSSI patterns, or payload tokens. [Technical] Vanhoef et al., 2016; Martin et al., 2017; Becker et al., 2019.
  • Secure key management. Secure element/TEE; access control; rotation; zeroisation; tamper detection. [Law] Art. 32.
  • Sensor hardening. Secure boot, signed firmware, secure updates, unique credentials, no default passwords, network segmentation, physical tamper protection. [Law] Art. 32.
  • No secondary use. Contractually prohibit re-identification, profiling, model training, sale, and sharing with third parties. [Law] Art. 28(3); [Guidance] EDPB 07/2020.

D. Transparency and rights

  • On-site signage before the measurement zone. Identify the controller, DPO, purposes, legal basis, data categories, retention, rights, opt-out, and complaint authority. [Law] Arts. 12–14, 21(4).
  • QR code to the full notice. German and English; accessible formats. [Guidance] CNIL, 24 Feb 2020 (two-level information).
  • Do not claim "fully anonymous" unless documented. Describe the staged processing precisely. [Guidance] EDPB 02/2026; [DPA] AP, "Wifi and bluetooth".
  • Right to object. Art. 21(4) notice; a functioning objection channel. [Law] Art. 21.
  • Opt-out mechanism. Dedicated Wi-Fi SSID wifi_tracking_optout for Wi-Fi; a web/app form for MAC/BLE; delete existing pseudonymous data on opt-out; suppress future matches; do not require turning off Wi-Fi/BLE as the only option. [Guidance] CNIL, 24 Feb 2020.
  • Acknowledge opt-out limits. MAC randomisation makes a MAC-based opt-out fragile; document the limitation and the alternative (spatial/temporal limits, no trajectories). [Technical] Martin et al., 2017.
  • Data subject request process. Art. 11/15–21; explain the limits; log requests; respond within one month. [Law] Arts. 11–12, 15–21.

E. Contracts and transfers

  • Art. 28 DPA with every processor. Scope, instructions, confidentiality, security, sub-processors, assistance, deletion/return, audit. [Law] Art. 28(3).
  • Art. 26 arrangement if joint controllers. Allocate transparency, rights, security, and breach responsibilities; publish the essence. [Law] Art. 26.
  • Sub-processor approval. LoRaWAN network operator, hosting provider, analytics provider; flow-down obligations; list of sub-processors. [Law] Art. 28(4).
  • EU hosting and transfers. Prefer EU/EEA; if outside, Art. 44–49 mechanism, SCCs + TIA, or adequacy decision; Schrems II. [CJEU] C-311/18.
  • No raw/pseudonymous data to the vendor. If the vendor needs data for support, provide counts or synthetic data; no re-identification. [Law] Art. 28(3); [Guidance] EDPB 07/2020.
  • Deletion and certification at the end. Return or delete; certify; audit. [Law] Art. 28(3)(g).

F. Operations and publication

  • Publish only coarse aggregates. Zone × hour/day; k ≥ 10/20; suppression; complementary suppression; DP noise; no per-sensor minute data; no repeat visits/trajectories. [Guidance] EDPB 02/2026; CNIL, 24 Feb 2020.
  • Document the disclosure-control method. k, ε, suppression rules, aggregation rule, and the re-identification assessment. [Guidance] EDPB 02/2026, paras 84–87.
  • No ad-hoc query interface over raw counts. Fixed tables; privacy budget; no differencing. [Technical] Dwork & Roth, 2014.
  • Monitor and reassess. New auxiliary data, new sensors, new purposes, and new technology can change the anonymity assessment. [Guidance] EDPB 02/2026; ICO, "Anonymisation".
  • Incident response. Art. 33/34; processor notification; sensor compromise; key compromise; data exfiltration. [Law] Arts. 33–34.
  • Training and accountability. Staff and volunteers handling data; confidentiality; data minimisation; no shadow copies. [Law] Art. 32.

G. Suggested pilot parameters

Parameter Suggested starting point Note
Sensor Passive BLE/Wi-Fi; no active requests; no payload The most defensible § 25 position
Raw MAC Memory only; ≤ 1 s; no logs/pcap/flash Minimisation; breach impact
Dedup ID HMAC-SHA-256; per-sensor, per-window key; ≥ 128-bit output Not a 16-bit ID; not a plain hash
Key rotation Every 15 min or 1 h; destroy old key Prevent cross-window linkage
Dedup window 15 min (default) or 1 h with justification Align with reporting interval
Cross-sensor linkage None No trajectories; no repeat visits
Uplink Counts only No identifiers leave the sensor
Publication Zone × hour/day; k ≥ 10 (20 for sensitive) No per-sensor 15-min data
DP Laplace/Gaussian; fixed tables; ε budget Additional layer, not a safe harbour
Retention Raw 0; pseudonymous ≤ window; aggregates 12–24 months Document and verify deletion
Signage Every entrance; DE/EN; QR; DPO; opt-out Before the measurement zone
Opt-out Dedicated SSID + web form; delete existing data Acknowledge MAC-randomisation limits
DPIA Before pilot; Art. 35(3)(c); DPO; Art. 36 if needed LfDI BW Muss-Liste Nos. 2/10 for private operators
Contracts Art. 28 DPA; Art. 26 if joint; EU hosting No secondary use; audit; deletion
Legal basis Public task (Art. 6(1)(e)) or, for a private operator only, legitimate interest (Art. 6(1)(f)) § 25 memo; no consent at scale

H. Red flags

  • "We hash the MAC, so it is anonymous." — Hashing is pseudonymisation; a short/unsalted hash is weak.
  • "We do not store MACs, so GDPR does not apply." — The sensor processes the MAC in memory; the collection stage needs a legal basis.
  • "The sign is the consent." — A sign is information, not consent. [CJEU] Planet49.
  • "People can opt out by turning off Wi-Fi." — The CNIL says this is not satisfactory; the ULD says it is practically impossible on some devices.
  • "We use a global salt." — A global salt links all devices across time/locations; use per-sensor, per-window keys.
  • "We keep hashed IDs for six months to measure repeat visits." — That is a movement profile; JCDecaux shows the problem; the purpose is incompatible with anonymisation.
  • "We share data with the vendor for benchmarking." — That is secondary use; the vendor may become a controller; Art. 28(10).
  • "We combine with camera/payment data to validate counts." — That creates linkability and may create Art. 9 data; avoid.
  • "We publish per-sensor 15-minute counts." — Small cells can identify individuals; coarsen/suppress.
  • "The Enschede fine was annulled, so Wi-Fi tracking is legal." — The fine was annulled on evidence/procedure, not on the merits; the legal question is unsettled.
  • "The city is a public body, so no fines." — TDDDG § 28(4) excludes TDDDG fines, but GDPR supervisory measures, complaints, and reputational risk remain.

8. Citations

All URLs were accessed on 14 November 2026. Guidance and enforcement decisions are persuasive or fact-specific; they are not binding law unless stated.

8.1 EU legislation

  1. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data (General Data Protection Regulation), OJ L 119, 4.5.2016, p. 1–88. https://eur-lex.europa.eu/eli/reg/2016/679/oj
  2. Directive 2002/58/EC of the European Parliament and of the Council of 12 July 2002 concerning the processing of personal data and the protection of privacy in the electronic communications sector (Directive on privacy and electronic communications), OJ L 201, 31.7.2002, p. 37–47, as amended by Directive 2009/136/EC, OJ L 337, 18.12.2009, p. 11–36. https://eur-lex.europa.eu/eli/dir/2002/58/oj

8.2 CJEU case law

  1. Case C-582/14, Patrick Breyer v Bundesrepublik Deutschland, ECLI:EU:C:2016:779, 19 October 2016. https://curia.europa.eu/juris/liste.jsf?num=C-582/14
  2. Case C-434/16, Peter Nowak v Data Protection Commissioner, ECLI:EU:C:2017:994, 20 December 2017.
  3. Case C-184/20, Vyriausioji tarnybinės etikos komisija, ECLI:EU:C:2022:601, 1 August 2022.
  4. Case C-319/22, Gesamtverband Autoteile-Handel eV v Scania CV AB, ECLI:EU:C:2023:857, 9 November 2023.
  5. Case C-413/23 P, European Data Protection Supervisor v Single Resolution Board, ECLI:EU:C:2025:645, 4 September 2025.
  6. Case C-673/17, Bundesverband der Verbraucherzentralen und Verbraucherverbände – Verbraucherzentrale Bundesverband eV v Planet49 GmbH, ECLI:EU:C:2019:801, 1 October 2019.
  7. Case C-604/22, IAB Europe v Gegevensbeschermingsautoriteit, ECLI:EU:C:2024:214, 7 March 2024.
  8. Case C-40/17, Fashion ID GmbH & Co. KG v Verbraucherzentrale NRW eV, ECLI:EU:C:2019:629, 29 July 2019.
  9. Case C-210/16, Unabhängiges Landeszentrum für Datenschutz Schleswig-Holstein v Wirtschaftsakademie Schleswig-Holstein GmbH, ECLI:EU:C:2018:388, 5 June 2018.
  10. Case C-708/18, Asociaţia de Proprietari bloc M5A Scara A v S.C. ANAF, ECLI:EU:C:2019:1063, 11 December 2019.
  11. Case C-621/22, Koninklijke Nederlandse Lawn Tennisbond (KNLTB) v Autoriteit Persoonsgegevens, ECLI:EU:C:2024:857, 4 October 2024.
  12. Case C-252/21, Meta Platforms Inc. v Bundeskartellamt, ECLI:EU:C:2023:537, 4 July 2023.
  13. Case C-311/18, Data Protection Commissioner v Facebook Ireland Ltd and Maximillian Schrems, ECLI:EU:C:2020:559, 16 July 2020.

8.3 EDPB and Article 29 Working Party

  1. EDPB, Guidelines 2/2023 on Technical Scope of Art. 5(3) of ePrivacy Directive, Version 2.0, adopted 7 October 2024. https://www.edpb.europa.eu/system/files/2024-10/edpb_guidelines_202302_technical_scope_art_53_eprivacydirective_v2_en_0.pdf
  2. EDPB, Guidelines 01/2025 on Pseudonymisation, adopted 16 January 2025, version for public consultation. https://www.edpb.europa.eu/system/files/2025-01/edpb_guidelines_202501_pseudonymisation_en.pdf
  3. EDPB, Guidelines 02/2026 on Anonymisation, Version 1.0, adopted 7 July 2026 for public consultation. https://www.edpb.europa.eu/system/files_en?file=2026-09%2Fedpb_guidelines_202602_anonymisation_v1_en_0.pdf
  4. EDPB, Guidelines 07/2020 on the concepts of controller and processor in the GDPR, Version 2.0, adopted 7 July 2021. https://www.edpb.europa.eu/our-work-tools/our-documents/guidelines/guidelines-072020-concepts-controller-and-processor-gdpr_en
  5. Article 29 Data Protection Working Party, Opinion 4/2007 on the concept of personal data, WP 136, 20 June 2007. https://ec.europa.eu/justice/article-29/documentation/opinion-recommendation/files/2007/wp136_en.pdf
  6. Article 29 Data Protection Working Party, Opinion 05/2014 on anonymisation techniques, WP 216, 10 April 2014. https://ec.europa.eu/justice/article-29/documentation/opinion-recommendation/files/2014/wp216_en.pdf
  7. Article 29 Data Protection Working Party, Guidelines on Data Protection Impact Assessment (DPIA) and determining whether processing is "likely to result in a high risk", WP 248 rev.01, 4 October 2017, endorsed by the EDPB. https://ec.europa.eu/newsroom/article29/items/611236
  8. Article 29 Data Protection Working Party, Opinion 9/2014 on the application of Directive 2002/58/EC to device fingerprinting, WP 219, 25 November 2014 (cited by EDPB Guidelines 2/2023 for the MAC-address/OS-API example). https://ec.europa.eu/justice/article-29/documentation/opinion-recommendation/files/2014/wp219_en.pdf

8.4 German law and guidance

  1. Telekommunikation-Digitale-Dienste-Datenschutz-Gesetz (TDDDG), §§ 2(2) No. 6, 25, 28, 29. https://www.gesetze-im-internet.de/ttdsg/__25.html; https://www.gesetze-im-internet.de/ttdsg/__28.html; https://www.gesetze-im-internet.de/ttdsg/__29.html
  2. Entwurf eines Gesetzes zur Regelung des Datenschutzes und des Schutzes der Privatsphäre in der Telekommunikation und bei Telemedien (TTDSG), BT-Drs. 19/27441, p. 38 (explanatory memorandum on § 24 TTDSG, now § 25 TDDDG). https://dserver.bundestag.de/btd/19/274/1927441.pdf
  3. Datenschutzkonferenz (DSK), Orientierungshilfe der Aufsichtsbehörden für Anbieter:innen von digitalen Diensten (OH Digitale Dienste), Version 1.2, November 2024, paras 22, 26. https://www.datenschutzkonferenz-online.de/media/oh/OH_Digitale_Dienste.pdf
  4. Landesbeauftragter für Datenschutz und die Informationsfreiheit Baden-Württemberg (LfDI BW), Liste von Verarbeitungsvorgängen nach Art. 35 Abs. 4 DS-GVO für den nicht-öffentlichen Bereich ("Muss-Liste"), 2018, Nos. 2 and 10. https://www.baden-wuerttemberg.datenschutz.de/wp-content/uploads/2018/05/Liste-von-Verarbeitungsvorg%C3%A4ngen-nach-Art.-35-Abs.-4-DS-GVO-LfDI-BW.pdf
  5. LfDI BW / Ministerium für Justiz und Migration Baden-Württemberg, Prüfung der Notwendigkeit einer Datenschutzfolgenabschätzung – Schwellenwertanalyse, 29 November 2024 (noting that there is no positive list for the public sector in Baden-Württemberg). https://jum.baden-wuerttemberg.de/fileadmin/redaktion/m-jum/intern/PDF/Migration/Erlasse_und_Anwendungshinweise/2024/29.11.2024_Anlage_01_Schwellwertanalyse.pdf
  6. Unabhängiges Landeszentrum für Datenschutz Schleswig-Holstein (ULD), Tätigkeitsbericht 2019, §5.4.8 "Offline-Tracking/Ortung von Mobiltelefonen in Fußgängerzone", pp. 97–99. https://www.datenschutzzentrum.de/tb/tb37/uld-37-taetigkeitsbericht-2019.pdf
  7. Der Hamburgische Beauftragte für Datenschutz und Informationsfreiheit (HmbBfDI), 25. Tätigkeitsbericht Datenschutz 2014/2015, §V.9 "Indoor-Ortung von Personen in Geschäften", pp. 141–143. https://datenschutz-hamburg.de/fileadmin/user_upload/HmbBfDI/Datenschutz/Taetigkeitsberichte_Datenschutz/Taetigkeitsberichte_PDF/25.TB_2014-2015.pdf
  8. Bayerisches Landesamt für Datenschutzaufsicht (BayLDA), Tätigkeitsbericht 2015/16, §3.10 "Offline-Tracking", pp. 30–31, and §22.10 "Offline-Tracking", p. 148. https://www.lda.bayern.de/media/baylda_report_07.pdf
  9. Bundesbeauftragte für den Datenschutz und die Informationsfreiheit (BfDI), 23. Tätigkeitsbericht 2009/2010, on WLAN scanning of access points (MAC address + SSID + signal strength + geocoordinates as potentially personal data; not a phone-tracking case). https://www.bfdi.bund.de/SharedDocs/Publikationen/Taetigkeitsberichte/TB_BfDI/23TB_09_10.pdf?__blob=publicationFile&v=6
  10. BfDI, Datenschutz-Folgenabschätzungen und Listen von Verarbeitungsvorgängen. https://www.bfdi.bund.de/DE/Fachthemen/Inhalte/Technik/Datenschutz-Folgenabschaetzungen.html
  11. Bundesministerium für Wohnen, Stadtentwicklung und Bauwesen / Innenstadtprogramm, Methoden zur Erfassung und Analyse von Passantenfrequenzen (policy methods report; describes MAC-based counting as "an sich anonym" while acknowledging de-anonymisation — not a legal assessment). https://www.innenstadtprogramm.bund.de/SharedDocs/downloads/Webs/ZIZ/DE/Erfahrungsbericht_Passantenfrequenz.pdf?__blob=publicationFile&v=2

8.5 Other EU/EEA supervisory authorities

  1. Commission Nationale de l'Informatique et des Libertés (CNIL), Dispositifs de mesure d'audience et de fréquentation dans des espaces accessibles au public: la CNIL rappelle les règles, 24 February 2020. https://www.cnil.fr/fr/dispositifs-de-mesure-daudience-et-de-frequentation-dans-des-espaces-accessibles-au-public-la-cnil
  2. Conseil d'État (France), JCDecaux France, no. 393714, 8 February 2017, ECLI:FR:CECHR:2017:393714.20170208. https://www.conseil-etat.fr/fr/arianeweb/CE/decision/2017-02-08/393714
  3. Autoriteit Persoonsgegevens (Netherlands), Wifi and bluetooth (theme page, last edited 9 April 2025). https://www.autoriteitpersoonsgegevens.nl/en/themes/internet-and-smart-devices/wifi-and-bluetooth
  4. Autoriteit Persoonsgegevens, Legal bases for WiFi tracking and Bluetooth tracking (last edited 23 May 2025). https://www.autoriteitpersoonsgegevens.nl/en/themes/internet-and-smart-devices/wifi-and-bluetooth/legal-bases-for-wifi-tracking-and-bluetooth-tracking
  5. Autoriteit Persoonsgegevens, Boete wifitracking Enschede, 29 April 2021, updated 29 July 2026. https://www.autoriteitpersoonsgegevens.nl/documenten/boete-wifitracking-enschede
  6. Rechtbank Overijssel, Enschede / Autoriteit Persoonsgegevens, ECLI:NL:RBOVE:2024:594, 2 February 2024. https://uitspraken.rechtspraak.nl/details?id=ECLI%3ANL%3ARBOVE%3A2024%3A594
  7. Afdeling bestuursrechtspraak van de Raad van State, Autoriteit Persoonsgegevens / College van B&W Enschede, ECLI:NL:RVS:2026:4403, 29 July 2026. https://www.raadvanstate.nl/@159747/202401622-1-a3

8.6 UK Information Commissioner's Office (persuasive, not binding in the EU)

  1. ICO, Anonymisation. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/
  2. ICO, Pseudonymisation. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/
  3. ICO, Privacy-enhancing technologies (PETs). https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/privacy-enhancing-technologies/
  4. ICO, Data protection impact assessments (DPIAs). https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/accountability-and-governance/data-protection-impact-assessments-dpias/
  5. ICO, Wi-Fi location analytics (2016; no longer available at the original ICO URL). Key recommendations summarised in Bird & Bird, "ICO publishes new guidance on Wi-Fi location analytics" (2016). https://www.twobirds.com/en/insights/2016/uk/ico-publishes-new-guidance-on-wifi-location-analytics

8.7 Technical and academic sources

  1. Code for Heilbronn e.V., Passantenzahlen messen mit LoRa, 18 September 2025. https://codeforheilbronn.de/blog/2025/09/passantenzahlen-messen-mit-lora/
  2. ESP32-Paxcounter v3.8.0, open-source repository; see also the project's legal note, https://paxcounter.org/legalnote/. https://github.com/cyberman54/ESP32-Paxcounter
  3. libpax, counting library extracted from Paxcounter. https://github.com/dbinfrago/libpax
  4. PulseCast project, Source-code audit: how the reference PAX counter actually counts, research note 09-paxcounter-code-audit.md, 14 November 2026.
  5. L. Sweeney, "k-anonymity: A model for protecting privacy", International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 10(5), 2002, pp. 557–570. DOI: 10.1142/S0218488502001648.
  6. C. Dwork and A. Roth, "The Algorithmic Foundations of Differential Privacy", Foundations and Trends in Theoretical Computer Science 9(3–4), 2014, pp. 211–407. DOI: 10.1561/0400000042.
  7. J. Martin, T. Mayberry, C. Donahue, L. Foppe, L. Brown, C. Riggins, E. C. Rye, and D. Brown, "A Study of MAC Address Randomization in Mobile Devices and When it Fails", Proceedings on Privacy Enhancing Technologies 2017(4), pp. 365–383. DOI: 10.1515/popets-2017-0054.
  8. J. K. Becker, D. Li, and D. Starobinski, "Tracking Anonymized Bluetooth Devices", Proceedings on Privacy Enhancing Technologies 2019(3), pp. 50–65. DOI: 10.2478/popets-2019-0036.
  9. M. Vanhoef, C. Matte, M. Cunche, L. S. Cardoso, and F. Piessens, "Why MAC Address Randomization is not Enough: An Analysis of Wi-Fi Network Discovery Mechanisms", Proceedings of the 11th ACM Asia Conference on Computer and Communications Security (AsiaCCS 2016), pp. 413–424. DOI: 10.1145/2897845.2897860.
  10. Apple, Wi-Fi privacy (Platform Security Guide). https://support.apple.com/guide/security/secb9cb3140c/web
  11. Android Open Source Project, MAC randomization. https://source.android.com/docs/core/connect/wifi-mac-randomization

End of note. This is a research note, not legal advice. Before deployment, obtain a written German law opinion on § 25 TDDDG, a DPIA sign-off, and a review of the actual firmware's hash length, key management, and cache behaviour.