# Bluetooth vs Wi-Fi: the OSI Model, TCP/IP, HTTP, FTP and How BLE Works

> Bluetooth vs Wi-Fi compared by protocol layer, not gadgets: the OSI and TCP/IP mapping, whether HTTP and FTP run over Bluetooth, and how BLE GATT and IPSP work.

- Published: 2026-11-17
- Author: 友田 陽大
- Tags: Bluetooth, IoT, 技術選定, アーキテクチャ設計, インフラ
- URL: https://tomodahinata.com/en/blog/bluetooth-vs-wifi-osi-tcp-ip-http-ftp-guide
- Category: Bluetooth, BLE & IoT integration

## Key points

- Wi-Fi (IEEE 802.11) defines the MAC and PHY of a wireless LAN, and ordinary IP, TCP and HTTP run on top of it. Bluetooth defines everything from the radio up to an application data model (GATT) and profiles in a single family of specifications, and whether it uses IP depends on how the stack is configured.
- Classic (BR/EDR) and BLE (LE) are separate, incompatible radio technologies; BLE is not a low-power mode of Classic. As of November 17, 2026, the latest version is Bluetooth Core Specification 7.0 (adopted November 2, 2026).
- Using HTTP directly over Bluetooth requires a configuration that carries IP. The ones whose specifications I could verify are PAN/BNEP on BR/EDR, and IPSP with RFC 7668 on LE. Bluetooth's "FTP" is the File Transfer Profile built on OBEX, which is unrelated to RFC 959 FTP on TCP port 21.
- Nominal PHY rate and application throughput are different things. By the author's calculation, even with the LE 2M PHY and Data Length Extension, the theoretical one-way upper limit is about 1.4 Mb/s. In the BLE 4.0 generation, the theoretical value was 236.7 kbps, while a 2012 measurement reached 58.48 kbps.
- A sensor that periodically sends small values is well served by GATT and needs no HTTP server on the device. Bulk transfer or direct reachability over IP favors Wi-Fi. This decision is summarized in a technology selection matrix.

---

If you design IoT devices, or you are a web engineer who wants to control a BLE device from the browser, the consumer rule of thumb ("earbuds use Bluetooth, the internet uses Wi-Fi") does not help you. What you actually need to know is this: can Bluetooth speak HTTP? Can it do FTP? Where does it sit in the OSI reference model?

This article is not a buyer's guide. It is **a comparison of protocol layers**. Specifications come from the Bluetooth Core Specification and RFCs, performance figures come from peer-reviewed papers, and anything that is my own calculation or judgment is labeled as such.

> **What you will learn**
> - How far up the stack Wi-Fi and Bluetooth each define things
> - Why Classic (BR/EDR) and BLE are separate radio technologies
> - When HTTP, TCP/IP and FTP work over Bluetooth, and when they do not
> - The theoretical upper limit of BLE application throughput (my own calculation), and how far measured values in papers fall below it
> - A decision table for choosing between a sensor, a gateway and a direct Wi-Fi connection

**The short answer**: Wi-Fi is "a wireless LAN for carrying IP," and everything above it is ordinary TCP/IP. Bluetooth is a family of specifications with its own stack, from the radio all the way up to the shape of application data (GATT). So the answer to "Can I use HTTP over Bluetooth?" is "Yes, but only in a configuration that carries IP." FTP is similar: RFC 959 FTP works once you have IP, but Bluetooth's "FTP" is something else entirely, the File Transfer Profile built on OBEX.

The specifications write Bluetooth Low Energy as "LE." In this article "BLE" and "LE" mean the same thing, and all version information is as of November 17, 2026.

---

## 1. The short version: one comparison table

The "Scope of the standard" row below is the subject of this whole article.

| Aspect | Wi-Fi (IEEE 802.11) | Bluetooth Classic (BR/EDR) | Bluetooth LE (BLE) |
| --- | --- | --- | --- |
| Standards body | IEEE (certification by Wi-Fi Alliance) | Bluetooth SIG | Bluetooth SIG |
| Scope of the standard | MAC and PHY of a wireless LAN | Radio up to profiles | Radio up to ATT, GATT and profiles |
| Relationship to IP | IP runs directly on top | Can carry IP via PAN/BNEP | Can carry IPv6 via IPSP and RFC 7668 |
| HTTP | Works as is | Works via PAN | Works via IPSP; not with GATT alone |
| Frequency bands | Mainly 2.4 / 5 / 6 GHz (also sub-GHz HaLow and 60 GHz WiGig) | 2.4 GHz ISM band | 2.4 GHz ISM band |
| Channels | Vary by generation and band | 79 channels, 1 MHz spacing | 40 channels, 2 MHz spacing |
| Nominal PHY rate | Varies widely by generation | 1 Mb/s; EDR 2 / 3 Mb/s | 1 Mb/s, 2 Mb/s, Coded 500 / 125 kb/s; 7.0 HDT up to 7.5 Mb/s |
| Current version | IEEE 802.11-2024 | Core Specification 7.0 | Core Specification 7.0 |

Terminology: **L2CAP** is the Bluetooth layer that multiplexes and segments upper-layer protocols. **PAN/BNEP** is the profile and protocol that carry the same packets as Ethernet over BR/EDR. **IPSP** is the profile for carrying IPv6 over LE. Each section gives the sources for its numbers.

---

## 2. Different goals: a wireless LAN beneath IP vs a short-range radio with its own stack

### 2-1. Wi-Fi is "a wireless LAN that carries IP"

Part 11 of IEEE 802.11-2024 is titled "Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications." The Wi-Fi standard defines how the signal is modulated (PHY) and who transmits when (MAC), and stops there.

IP runs on top. In the four-layer TCP/IP model, Wi-Fi belongs to the same "link layer" as Ethernet.

### 2-2. Bluetooth goes all the way up to "the shape of application data"

In addition to the radio, the link layer and L2CAP, the Bluetooth Core Specification defines ATT (Attribute Protocol) and GATT (Generic Attribute Profile). GATT describes the data a device holds, and how to read and write it, in units called "services" and "characteristics." An operation such as reading a heart-rate monitor's value completes without IP. On top of that, the Bluetooth SIG publishes "profiles" as separate specifications, such as PAN and the File Transfer Profile. A profile is a way of combining the stack for a specific use case.

So for a Wi-Fi device, communication design is a question of "which IP protocol do I use?" For a Bluetooth device, you first have to decide "do I use IP at all, or is GATT enough?"

---

## 3. Physical and link layers: how each shares the 2.4 GHz band

### 3-1. They share the same band

Both BR/EDR and LE operate in the unlicensed 2.4 GHz ISM band. The Core Specification notes that this band is shared with "wireless LAN, cordless phones, and microwave ovens." Wi-Fi also uses the 5 GHz and 6 GHz bands (the 802.11be-2024 overview lists coexistence with legacy devices in the 2.4 / 5 / 6 GHz bands), and the Wi-Fi Alliance also lists Wi-Fi HaLow in sub-GHz bands and WiGig in the 60 GHz band.

### 3-2. Bluetooth channels

Core Specification 7.0 defines them as follows.

- **BR/EDR**: 79 channels at 1 MHz spacing. Center frequencies are `2402 + k MHz` (k = 0 to 78).
- **LE**: 40 channels at 2 MHz spacing, from 2402 to 2480 MHz. Three of them (2402 / 2426 / 2480 MHz) are the "primary advertising channels," and the remaining 37 are general-purpose channels.

Advertising, the way a BLE device announces its presence, starts on these three channels (since 5.0 there is also secondary advertising on the general-purpose channels). Gomez et al. (2012) explain that the three channels are placed to minimize overlap with 802.11 channels 1, 6 and 11.

### 3-3. Who transmits when: a coordinator vs taking turns

The BR/EDR Baseband specification calls a group of devices sharing a channel a "piconet," in which one device is the Central and the rest are Peripherals. The specification states that a piconet can have at most seven active Peripherals and that the Central controls access to the channel. In LE, communication over a connection likewise happens in units called "connection events," at times determined by the connection parameters.

"Bluetooth supports up to 7 devices" refers to this piconet. The number of simultaneous LE connections is determined by the chip and the stack's implementation and configuration. For example, NimBLE-Arduino 2.5.2 defaults to 3, and a comment in its source gives 9 as the limit of the ESP32 controller.

Wi-Fi is distributed. Ferro and Potortì's 2005 survey describes DCF, the basic 802.11 MAC, as a CSMA/CA channel access scheme. A station listens before transmitting to check whether the medium is idle; if it is busy, the station waits a random backoff time before sending. According to the same survey, a piconet's aggregate throughput does not depend on the amount of traffic flowing. In a Wi-Fi BSS (one access point and the stations associated with it), by contrast, efficiency drops as load increases.

My own reading is that Bluetooth, where "the Central decides the order," suits carrying small data on schedule, while Wi-Fi, where "you send when the medium is free," suits streaming large data. Note, however, that the survey compares against the Bluetooth 1.2 era, and real performance depends on congestion and implementation.

---

## 4. Classic (BR/EDR) and LE are different radio technologies

### 4-1. They are not compatible

The Core Specification's Architecture volume states: "There are two forms of Bluetooth wireless technology systems: Basic Rate (BR) and Low Energy (LE)." EDR (Enhanced Data Rate) is an extension of BR. RFC 7668 is even more explicit: "Bluetooth and Bluetooth LE are different, non-interoperable radio technologies."

A "dual-mode" device that implements both can talk to either kind, but a BR/EDR-only device and an LE-only device cannot communicate with each other.

### 4-2. BLE is not "a low-power mode of Classic"

A common misconception is that "BLE is Classic's low-power mode." In reality, almost everything differs: the number and spacing of channels, the hopping scheme (Section 12), modulation and data rates (Section 13), and the main upper-layer components. BR/EDR centers on RFCOMM, which emulates a serial port, on SDP (Service Discovery Protocol) for finding services, and on profiles; LE centers on ATT and GATT. LE is a separate radio technology, added in Bluetooth 4.0 (adopted June 30, 2010) together with its own PHY, link layer, ATT, GATT and Security Manager.

### 4-3. The 3.0+HS "borrow Wi-Fi" mechanism is gone

Version 3.0+HS (adopted April 21, 2009) introduced the Alternative MAC/PHY (including the 802.11 PAL), which sent large data over an 802.11 radio. It was removed in Core 5.3. "Bluetooth can use Wi-Fi to go faster" does not apply to the current specification.

### 4-4. The current version is Core Specification 7.0

As of November 17, 2026, the latest version is Core Specification **7.0**, adopted on November 2, 2026. After 6.0 (adopted August 27, 2024) came 6.1 (May 5, 2025), 6.2 (November 3, 2025) and 6.3 (May 4, 2026). Statements that "the latest is 6.0" are already out of date.

---

## 5. Mapping to the OSI reference model and the TCP/IP model

A table mapping the seven OSI layers to the four TCP/IP layers is in the [TCP/IP complete guide](/blog/tcp-ip-protocol-suite-fundamentals-complete-guide). The key point is that TCP/IP is defined in RFC 1122 as four layers (Application / Transport / Internet / Link), whereas the seven-layer OSI model is a conceptual framework for organizing ideas.

### 5-1. Placing Bluetooth in OSI is an approximation

The Bluetooth Core Specification is not written in terms of OSI layers. It is defined as a **Controller**, which handles the radio and link layer, and a **Host**, which handles L2CAP and above, connected by HCI (Host Controller Interface). HCI itself is optional.

Some profile specifications do mention OSI, though.

- The PAN profile describes Baseband, LMP and L2CAP as "the part of the Bluetooth protocols that reside in the OSI layers 1 and 2." LMP (Link Manager Protocol) is the protocol that sets up and manages BR/EDR links.
- The BNEP specification says that "Bluetooth is considered to be a transmission media in the same OSI layer as Ethernet, Token Ring, ATM, etc.," and goes on to treat L2CAP as the "Bluetooth Data MAC (Media Access Control) Layer."

From the point of view of carrying IP, Bluetooth is a "layer 1-2 medium," just like Ethernet or Wi-Fi. From the point of view of an application that stays within GATT, ATT and GATT also play a role close to the application layer. That is why **Bluetooth as a whole cannot be squeezed into a single OSI layer**. The mapping in this article is also a conceptual approximation.

### 5-2. Four stacks side by side

The diagram below shows four stacks, bottom to top. In Wi-Fi, IP sits on top of the 802.11 PHY and MAC. In BR/EDR+PAN, BNEP sits on L2CAP and IP sits on BNEP. In LE+IPSP, 6LoWPAN header compression sits on L2CAP and IPv6 on top of that. In these three, TCP or UDP and then HTTP sit on top of IP. LE+GATT alone contains no IP: ATT, GATT and application data sit directly on L2CAP.

```text
 Wi-Fi       BR/EDR+PAN    LE+GATT      LE+IPSP
+-----------+------------+------------+------------+
| HTTP etc. | HTTP etc.  | app data   | HTTP etc.  |
| TCP / UDP | TCP / UDP  | GATT       | TCP / UDP  |
| IP        | IP         | ATT        | IPv6       |
|           | BNEP       |            | 6LoWPAN    |
|           |            |            | (IPHC)     |
| 802.11    | L2CAP      | L2CAP      | L2CAP      |
| MAC       |            | (CID 0x04) | (LE CoC)   |
|           | LMP/       | Link Layer | Link Layer |
|           | Baseband   |            |            |
| 802.11    | BR/EDR     | LE PHY     | LE PHY     |
| PHY       | Radio      |            |            |
+-----------+------------+------------+------------+
  IP: yes     IP: yes      IP: no       IP: yes
```

The diagram is simplified (layers between the Wi-Fi MAC and IP, HCI, GAP and SMP are omitted). Whether "Bluetooth can use IP" is decided by how the stack is assembled, not by the type of radio.

---

## 6. The BLE stack: Link Layer, L2CAP, ATT, GATT (plus SMP and GAP)

### 6-1. Link Layer and L2CAP

An LE 1M / 2M **Link Layer** packet consists of a preamble (1 octet on 1M, 2 octets on 2M), a 4-octet access address, a PDU of 2 to 258 octets, and a 3-octet CRC. The payload of a data PDU is at most 27 octets without Data Length Extension (DLE), or at most 251 octets with DLE, introduced in Bluetooth 4.2. On an encrypted connection, a 4-octet MIC (Message Integrity Check) for tamper detection follows the payload.

The basic **L2CAP** header is 4 octets: a 2-octet length and a 2-octet channel ID. On LE, ATT uses the fixed channel ID 0x0004.

### 6-2. ATT and GATT

**ATT** is a client-server protocol for reading and writing the "attributes" a peer device holds, addressed by numbers called handles. **GATT** is the set of rules that organizes those attributes into a hierarchy of "services" (a collection of data and behavior for a particular function) and "characteristics" (a value within a service, plus information on how to access it). There are four main operations for delivering values.

| Operation | Direction | Acknowledgment | Key point in the spec |
| --- | --- | --- | --- |
| Write Request | Client → Server | Write Response returned | Server responds after writing |
| Write Command | Client → Server | None | Returns neither a response nor an error |
| Notification | Server → Client | None | No acknowledgment PDU is defined |
| Indication | Server → Client | Confirmation returned | No further Indication until the Confirmation arrives |

A client that wants to receive updates sets bits in a descriptor called the Client Characteristic Configuration (UUID 0x2902). The GATT specification says that when both the Notification and Indication bits are set, the server uses Notification when no acknowledgment is needed and Indication when one is.

### 6-3. Where "BLE is limited to 20 bytes" comes from

"BLE can send at most 20 bytes at a time" is not a limit in the specification.

- The default ATT_MTU on LE is **23 octets** (GATT specification).
- A Write or Notification PDU carries a 1-octet opcode and a 2-octet handle before the value, so one operation can carry `ATT_MTU - 3` octets of value.
- With the default, that is `23 - 3 = 20`. This is "the value that works with any peer without negotiating the MTU."

ATT_MTU can be negotiated after connecting, and it settles on the smaller of the two sides' receive capabilities. For example, with an ATT_MTU of 247 (an example for the calculation, not a constant in the specification) and a 251-octet DLE payload, one packet can carry a `247 - 3 = 244`-octet value (251 = 4 L2CAP header + 3 ATT header + 244 value). The maximum length of an attribute value itself is **512 octets**.

The Web Bluetooth API does not expose the negotiated MTU and rejects writes longer than 512 octets (see the [Web Bluetooth API implementation guide](/blog/web-bluetooth-api-javascript-typescript-ble-guide)).

### 6-4. SMP and GAP

**SMP (Security Manager Protocol)** generates encryption keys and identity keys and runs over its own fixed L2CAP channel. **GAP (Generic Access Profile)** defines the basic behavior common to all Bluetooth devices, such as discovery and connection modes.

---

## 7. Can you use HTTP over Bluetooth?

**Answer: yes, but only in a configuration that carries IP.**

RFC 9110 notes that "HTTP is independent of the transport protocol," yet it defines the `http` and `https` schemes in terms of TCP connections (default ports 80 and 443). HTTP/3 uses QUIC over UDP (RFC 9114). Every version needs IP underneath.

I was able to verify the specifications of two ways to carry IP over Bluetooth (the IP Link Protocol in 9-3 is unverified in detail):

1. **The PAN profile and BNEP on BR/EDR** (Section 8)
2. **IPSP and RFC 7668 on LE** (Section 9)

A BLE device that communicates only through GATT has no IP address and no ports. If a browser or the cloud needs to talk to such a device over HTTP, you put "something that speaks both GATT and IP" in between to translate: a smartphone app, a gateway, or a web page using Web Bluetooth.

This translation has also been standardized by the Bluetooth SIG as the HTTP Proxy Service (HPS) 1.0. HPS is a service that lets a client device, typically a sensor, communicate with a web server through a gateway. The sensor writes over GATT to the gateway's URI and header characteristics, and the gateway performs the actual HTTP exchange (the sensor itself has no IP).

TCP (RFC 9293) also runs over IP. For more, see [the TCP three-way handshake and state transitions](/blog/tcp-three-way-handshake-state-transition-retransmission-congestion-control-guide) and [TCP vs UDP vs QUIC](/blog/tcp-vs-udp-quic-http3-difference-when-to-use-guide).

---

## 8. Carrying IP over Classic: PAN and BNEP

BNEP (Bluetooth Network Encapsulation Protocol) 1.0 is a specification for carrying, directly over L2CAP, the same network protocols that IEEE 802.3 (Ethernet) encapsulation can carry. It is sometimes described as "carrying Ethernet frames as they are," but the specification actually says: "BNEP removes and replaces the Ethernet Header with the BNEP Header." BNEP requires an L2CAP MTU of at least 1691 octets, and its identifier on L2CAP (PSM) is 0x000F.

PAN (Personal Area Networking Profile) 1.0 is the profile that defines how to build networks with BNEP. It has three roles: NAP (Network Access Point), which has part of the functionality of an Ethernet bridge and forwards Ethernet packets between connected devices; GN (Group Ad-hoc Network), which forwards packets among devices; and PANU (PAN User), the devices that connect to them. Through a NAP, a PANU device can reach the LAN or the internet beyond the NAP over IP. PAN is a profile that runs over BR/EDR; PAN 1.0 has no LE transport.

PAN 1.0 and BNEP 1.0 were both adopted on February 14, 2003. In the specification list as of November 17, 2026, both were "Adopted," with no deprecation date listed. PAN support is decided product by product, so check first which roles the peer device and its OS support.

---

## 9. Carrying IP over LE: IPSP, RFC 7668 and RFC 9159

### 9-1. IPSP and RFC 7668

The standard way to carry IP over LE is the combination of the Bluetooth SIG's IPSP (Internet Protocol Support Profile) 1.0 and the IETF's RFC 7668, "IPv6 over BLUETOOTH(R) Low Energy" (October 2015).

- RFC 7668 assumes Bluetooth 4.1 or later and IPSP 1.0 or later.
- IPSP specifies that "only the LE Connection Oriented Channels feature with the LE Credit Based Flow Control Mode shall be used" to carry data, with an MTU of at least 1280 octets. Its identifier on L2CAP (SPSM) is 0x0023, and it has two roles, Node and Router.
- RFC 7668 makes RFC 6282 mandatory for IPv6 header compression. 6LoWPAN is a family of IETF techniques for carrying IPv6 over low-rate links with small frames, and IPHC is its header compression format.

The reason for 1280 is that the IPv6 specification, RFC 8200, requires that "every link in the Internet have an MTU of 1280 octets or greater." Because the LE link-layer payload is at most 251 octets, RFC 7668 recommends an MTU of 1280 octets and leaves fragmentation and reassembly to L2CAP.

### 9-2. RFC 9159: an IPv6 mesh

RFC 9159 (December 2021) defines how to build an IPv6 **mesh** over LE using IPSP. It mandates a "route-over" approach, in which routing decisions are made at the IP layer, and does not specify a routing protocol. It is unrelated to the Bluetooth SIG's "Bluetooth Mesh."

Note that one sentence in RFC 9159, "Bluetooth 4.2 defines a default MTU for Bluetooth LE of 251 bytes," conflicts with the Core Specification. 251 is the **maximum** link-layer payload with DLE; the default is 27 octets.

### 9-3. IP Link Protocol (details unverified)

The Bluetooth SIG website has a specification page titled "IP Link Protocol" (the page shows a date of August 6, 2026), which says that it "specifies how Internet Protocol (IP) Version 6 (IPv6) packets are transported over Bluetooth® Low Energy (LE)." Assigned Numbers also lists SPSM 0x0029 (IP_LINK). However, I could not retrieve the specification PDF, and its relationship to IPSP and RFC 7668 is **unverified as of November 17, 2026**.

### 9-4. Comparing the data flows

The diagram below shows, top to bottom, how a value is wrapped on its way out in three configurations. (A) Over Wi-Fi, an HTTP request is wrapped in TCP, IP and an 802.11 frame, and leaves through the access point to the LAN. (B) Over Bluetooth PAN, below IP it is wrapped in BNEP and L2CAP and sent over BR/EDR, and the NAP forwards it to the LAN as an Ethernet bridge. (C) Over a direct GATT link, the value becomes an ATT Write Command or Notification, is wrapped in L2CAP and a link-layer PDU, and goes straight to the peer device without IP, TCP or HTTP.

```text
 (A) Wi-Fi        (B) BT PAN        (C) LE GATT
 -------------    --------------    ---------------
 HTTP request     HTTP request      value (bytes)
   |                |                 |
 TCP segment      TCP segment       ATT Write Cmd
   |                |               or Notification
 IP packet        IP packet           |
   |                |               L2CAP
 802.11 frame     BNEP header       (CID 0x0004)
   |                |                 |
   |              L2CAP             LL data PDU
   |                |                 |
 [802.11 radio]   [BR/EDR radio]    [LE radio]
   |                |                 |
   v                v                 v
 AP -> LAN/WAN    NAP -> LAN        peer device
                  (Ethernet         (no IP stack)
                   bridge)
```

(A) and (B) differ only in the link layer; to the application, both are ordinary IP networks. In (C), the application handles GATT operations directly.

---

## 10. FTP and "Bluetooth FTP" share a name and nothing else

In RFC 959 FTP (October 1985), the control connection is TCP, on server port 21 ("service port 21 (25 octal)"). Files travel only over a separate data connection, and the default data port is written as "L-1." "Port 20" is derived from `21 - 1`; the number itself does not appear in the RFC.

Bluetooth's "FTP" stands for **File Transfer Profile**. The current version is 1.3.1 (December 15, 2015), and it uses GOEP (Generic Object Exchange Profile) and OBEX. OBEX is "the Bluetooth adaptation of IrDA Object Exchange Protocol (IrOBEX)"; it runs over either L2CAP or RFCOMM, with BR/EDR underneath. The Object Push Profile (currently 1.2.1) also uses OBEX. Their SDP identifiers are 0x1105 for OBEX Object Push and 0x1106 for OBEX File Transfer.

For moving large data over LE, the relevant SIG specification is not the File Transfer Profile but the Object Transfer Service (OTS) 1.0. OTS is a GATT service, and the object contents travel over a separate L2CAP connection-oriented channel (per the specification, that channel can be used on both LE and BR/EDR).

If you want RFC 959 FTP over Bluetooth, set up IP with PAN or IPSP first and run it on top.

---

## 11. Wi-Fi radio techniques: OFDM, OFDMA and MIMO

This section is written without reading the IEEE 802.11 text itself; it relies on the standard pages' overviews, Wi-Fi Alliance public pages (which are not the normative text of the standard), and papers. According to Ferro and Potortì (2005), 802.11a/g OFDM transmits in parallel over 52 subcarriers (48 for data) spaced 0.3125 MHz apart, a different approach from Bluetooth's frequency hopping (FHSS). These are 802.11a/g values; Wi-Fi 6 and later differ. The Wi-Fi Alliance lists OFDMA (several stations sharing a channel), multi-user MIMO and TWT (Target Wake Time, which it promotes as improving battery life) as Wi-Fi 6 features, and 320 MHz channels **in the 6 GHz band only** and multi-link operation as Wi-Fi 7 features.

IEEE 802.11-2024 was approved on September 26, 2024 and published on April 28, 2025. 802.11be-2024 (Wi-Fi 7) was approved on the same day and **published on July 22, 2025**. 802.11be's target of "at least 30 Gbit/s, as measured at the MAC data service access point (SAP)" (the interface to upper layers) is a design goal, not a device speed.

---

## 12. Frequency hopping and AFH

### 12-1. BR/EDR: 1600 hops per second

BR/EDR hops among its 79 channels in a pseudo-random order. The maximum hop rate is 1600 hops per second in a connection and 3200 hops per second during Inquiry and Page. 1600 hops per second means 625 µs per hop. The specification keeps the frequency fixed for the duration of a packet, so multi-slot packets reduce the number of hops.

### 12-2. LE: hopping per connection event

LE switches channels **at the start of each connection event**, not per packet, and stays on the same channel for the whole connection event. The first channel selection algorithm (CSA #1) uses the formula below; if it lands on an unused channel, it remaps to a used one.

```text
unmappedChannel =
    (lastUnmappedChannel + hopIncrement) mod 37
hopIncrement : random value in 5..16, set at connection
```

Bluetooth 5.0 added CSA #2, which uses a pseudo-random number generator.

### 12-3. AFH: avoiding busy channels

AFH (Adaptive Frequency Hopping) removes channels with interference from the hopping set. On LE, AFH support on the general-purpose channels is **mandatory**, and channels are classified as "unknown," "bad" or "good." BR/EDR can also exclude, per Peripheral, the frequencies used by interfering devices. The Core Specification says the adaptive hopping technique "improves Bluetooth co-existence with static (non-hopping) ISM systems" such as Wi-Fi.

It has limits, though. Spörk et al. (2019, Section 14) pointed out that an LE Peripheral can detect bad channels but has no standard way to update the channel map. That reflects the specification at the time of the paper: Core 5.3 added LE Channel Classification, which lets a Peripheral report channel quality to the Central (per the Architecture volume, the Peripheral can classify frequencies as good or bad and provide that information to the Central). The Central is still the one that updates the map, and this feature is optional, so support needs to be checked product by product (as of November 17, 2026, I have not verified which products support it).

---

## 13. Speed, latency and power: separate nominal from effective

### 13-1. Nominal PHY rates

The rates in the specifications are "how many bits per second go over the air," not the rate at which an application receives data.

| Technology | Value in the spec | Kind |
| --- | --- | --- |
| BR | 1 Mb/s | PHY bit rate |
| EDR | 2 Mb/s, 3 Mb/s | Gross air bit rate |
| BR / EDR data rate | 721.2 kb/s, 2.1 Mb/s | Data rate given in the spec's Architecture volume |
| LE 1M | 1 Mb/s (mandatory) | PHY bit rate |
| LE 2M | 2 Mb/s (from 5.0, optional) | PHY bit rate |
| LE Coded | 500 kb/s at S=2, 125 kb/s at S=8 | PHY bit rate |
| LE HDT | 2 / 3 / 4 / 6 / 7.5 Mb/s (from 7.0, optional) | PHY bit rate |

Wi-Fi is no different. Ferro and Potortì (2005) calculate that even at 802.11g's 54 Mb/s, with 1500-byte packets and no collisions or interference, the maximum one-way throughput is 31.4 Mb/s with CSMA/CA and 26.7 Mb/s with RTS/CTS.

### 13-2. My calculation: the theoretical upper limit of LE application throughput

What follows is **my own calculation, not a measurement**. From the packet formats in Core Specification 7.0, I derive the theoretical upper limit when one side keeps sending values in one direction. The assumptions are:

- The sender transmits data PDUs and the receiver answers with Empty PDUs, repeated without limit within a connection event.
- The inter-frame space (T_IFS) is the default 150 µs. No packet loss.
- ATT uses Write Command or Notification, which do not wait for a response (Write Request is much slower).
- The application value is 20 octets without DLE and 244 octets with DLE (ATT_MTU 247). With encryption, each data PDU carries a 4-octet MIC.

```text
airtime = (preamble + 4 + 2 + payload + MIC + 3)
          x 8 us (LE 1M) / x 4 us (LE 2M)
  preamble: 1 (1M) / 2 (2M), MIC: 0 or 4 [octets]
period  = data PDU + T_IFS + empty PDU + T_IFS
limit   = bits of app value / period

Example: LE 2M + DLE, unencrypted
  data PDU  = (2 + 4 + 2 + 251 + 0 + 3) x 4 = 1048 us
  empty PDU = (2 + 4 + 2 +   0 + 0 + 3) x 4 =   44 us
  period    = 1048 + 150 + 44 + 150         = 1392 us
  limit     = 244 x 8 / 1392 = 1952 / 1392  = 1.402 Mb/s
```

Here are the results for five conditions. I checked them with a script and confirmed that they agree with the 328 µs and 2120 µs (MIC included) in the specification's tables and with the 44 µs minimum packet time.

| Condition | Period | Theoretical upper limit |
| --- | --- | --- |
| LE 1M, no DLE (20-octet value), unencrypted | 676 µs | 0.237 Mb/s |
| LE 1M, DLE (244-octet value), unencrypted | 2468 µs | 0.791 Mb/s |
| LE 1M, DLE, encrypted | 2500 µs | 0.781 Mb/s |
| LE 2M, DLE, unencrypted | 1392 µs | 1.402 Mb/s |
| LE 2M, DLE, encrypted | 1408 µs | 1.386 Mb/s |

Even with a 2 Mb/s PHY, headers, the CRC, inter-frame spaces and the peer's response time bring the upper limit down to about 1.4 Mb/s. "BLE 2M gets you 2 Mbps" confuses the PHY rate with effective throughput. The 0.237 Mb/s in the first row matches the 236.7 kbps that Gomez et al. and Tosi et al. give as the theoretical value under BLE 4.0 conditions, because it is computed with the same model.

### 13-3. What the papers measured, and under what conditions

Real devices do not reach this upper limit. The main reason is that stacks limit how many packets can be sent per connection event.

- **Gomez et al. (2012, Bluetooth 4.0)**: with two TI CC2540s 0.5 m apart, a 7.5 ms connection interval and 20-octet Notifications, the maximum application-layer throughput measured was **58.48 kbps**. Because of a stack limitation, only 4 packets could be sent per connection event (11 in theory).
- **Tosi et al. (2017, systematic review)**: they conclude that while about 230 kbps is theoretically achievable, about 100 kbps was the ceiling in the real-world applications they reviewed. They report that many Android devices send about 4 packets per connection interval, many iOS devices at most 6, and that the shortest connection interval is 20 ms.

When the packet count is capped, the upper limit is `packets x bits per value / connection interval`. Under Gomez et al.'s conditions that is `4 x 160 / 7.5 ms = 85.3 kb/s`, and the measured 58.48 kbps was lower than that (my calculation).

These are figures from the BLE 4.x generation, before DLE (4.2) and the LE 2M PHY (5.0), and must not be quoted as today's limits. Conversely, you also cannot say that "newer devices come close to the theoretical limit"; the per-event packet cap is still implementation-dependent.

### 13-4. How newer specifications change the limit

Newer versions include optional features that change the assumptions of the calculation above. They work only when both devices support them, and as of November 17, 2026, I have not verified any products that support them.

- **Frame Space Update (6.0)**: T_IFS can be changed per connection; if it gets shorter, the "period" shrinks.
- **Shorter Connection Intervals (6.2)**: connection intervals range from 375 µs to 4 s in 125 µs steps; controllers that do not support the feature continue to use only values of 7.5 ms or more. In my analysis, on implementations where the packet cap applies per connection event, shorter intervals may raise the limit, but they do not exceed the continuous-transmission limit in 13-2.
- **LE HDT PHY (7.0)**: it combines three modulations (π/4 QPSK, 8PSK and 16QAM) with coding to provide five data rates: 2 / 3 / 4 / 6 / 7.5 Mb/s. The payload field can be up to 494 octets (8175 if both sides support it), but the specification notes that the total payload that fits in the 3357 µs air-time limit is less than 8175 bytes.

So "Bluetooth 7 is 7.5 Mbps" is inaccurate. 7.5 Mb/s is the maximum bit rate of the optional HDT PHY; it is neither application throughput nor the speed of every device.

### 13-5. Latency and power

- **Latency**: Gomez et al. measured the latency of an error-free one-way ATT exchange on the CC2540 at **676.7 µs**. The paper explains that this corresponds to sending a maximum-length Notification, receiving the link-layer acknowledgment, and waiting two inter-frame spaces. It does not include waiting for the next connection event. In my analysis, the application sees up to roughly one extra connection interval of waiting in the worst case, so longer connection intervals save power at the cost of higher latency.
- **Power**: the typical currents in Ferro and Potortì (2005), about 1-35 mA for Bluetooth and 100-350 mA for Wi-Fi, are Bluetooth 1.2-era values, and they are not energy per bit either. Gomez et al. estimated the lifetime of a BLE Peripheral running on a 230 mAh coin cell with a theoretical calculation using CC2540 current figures (ideal battery, no bit errors, one-way communication): at most 14.1 years (connection interval 86.25 ms, Peripheral latency 370; the paper calls it slave latency) and as little as 2.6 days (connection interval 7.5 ms, latency 0). For two-way communication, the figures are at most 12.4 years and as little as 2.0 days. The point is how much the configuration changes the result.

---

## 14. Wi-Fi interference: does BLE "drop" or "lag"?

Since both use the 2.4 GHz band, Wi-Fi can interfere with BLE. Two experiments show that how the interference shows up depends on the conditions.

### 14-1. An experiment where throughput dropped sharply

Pang et al. (2022) placed two nRF52840 evaluation boards 40 cm apart in an office, using the LE 1M PHY and a 7.5 ms connection interval, with a strong Wi-Fi interferer.

- No interference: packet delivery ratio of 95-100%.
- Fixed interference on Wi-Fi channel 1: increasing the payload from 100 to 200 octets lowered the delivery ratio from 90% to 80%.
- At a 100-octet payload, the delivery ratio fell only **7.6 percentage points**, from 98.1% to 90.5%, but throughput fell **31.1%**, from 120,967 bps to 83,317 bps.

The authors attribute this to the retransmission mechanism, but it was not proven (they write that further research is needed). The measurements cover specific devices, a specific distance and one office.

### 14-2. An experiment where packets were delayed, not lost

Spörk et al. (EWSN 2019) used an nRF52840 as the Peripheral (the paper's term is slave) and a Raspberry Pi 3 as the Central (master), and measured latency over IPv6 over BLE. The connection interval was 250 ms, and the expected latency bound was 260 ms.

- Interferer **near the Central**: AFH excluded data channels 18 to 31, and only a few packets were delayed.
- Interferer **near the Peripheral**: the Central did not update the channel map, and some packets were delayed by more than 5 times the expected bound (5 × 260 ms = 1.3 seconds, my conversion). The same happened for every Peripheral 7 m or more away from the Central.
- Even so, every UDP packet **eventually arrived**. The interference showed up as delay, not loss.

The cause is the channel-map update problem mentioned in 12-3, and it assumes the specification at the time of the paper (see 12-3 for LE Channel Classification in 5.3). The authors' proposed scheme, which adjusts the connection interval, reduced delayed packets by up to a factor of 40, but it is not a standard BLE feature.

In other words, "resistant to interference" or "vulnerable to interference" cannot be decided in a word: depending on the conditions, throughput may drop or latency may grow. For products that carry both Wi-Fi and Bluetooth, the Core Specification says that "time division duplexing (TDD) techniques need to be used" when the radios are not sufficiently isolated or share an antenna.

---

## 15. Is a GATT design without TCP/IP reasonable?

This section is **my own design analysis**.

### 15-1. For small periodic values, the device does not need an HTTP server

If a temperature sensor sends a few octets every few seconds, GATT only requires placing the value in a characteristic and sending a Notification, and if it fits in 20 octets there is no need to negotiate the MTU.

Doing the same over HTTP requires IP address configuration, TCP, HTTP parsing and, in production, TLS, attaching many times more headers and handshakes to a value of a few octets. If the device acts as an HTTP server, you also have to solve who knows its address and how they reach it.

On the other hand, choosing GATT costs you the following:

- Reaching the internet requires a relay such as a smartphone or gateway.
- HTTP's rich tooling (caching, proxies, authentication schemes, monitoring) cannot be used as is.
- Using the device directly from a browser requires Web Bluetooth, and browser support is limited (Section 17).

### 15-2. For bulk transfer, BLE can lose to Wi-Fi

Energy is power multiplied by time. Even though Wi-Fi draws more power while transmitting, its energy per bit can be lower if it finishes quickly and goes back to sleep. BLE draws less power, but its effective throughput can be limited to tens or hundreds of kb/s (13-3), so for large files it keeps the radio active for a long time.

In the abstract of their smartphone measurements, Friedman et al. (2011 / 2013) state that the study "identifies many situations in which WiFi is superior to Bluetooth" and "a couple of scenarios that are better handled by Bluetooth." That result may be consistent with this direction (this is based on the abstract only; I have not verified which metric Wi-Fi was better on, or whether they measured BLE).

Conversely, if you only send small values occasionally, Wi-Fi's fixed costs, such as maintaining the connection and the IP and TLS handshakes, start to dominate. BLE lets you tune idle behavior finely with the connection interval and Peripheral latency (13-5).

My judgment is that "BLE for small, intermittent data; Wi-Fi for large, continuous data" is a starting point, but you will not know the power consumption until you measure it on real hardware.

---

## 16. Technology selection matrix

From everything above, I summarized a decision table for early-stage design, based on my own evaluation. "Best" means well suited, "Yes" means possible under conditions, "Extra work" means it takes extra work, and "No" means not suited.

| Requirement | Wi-Fi (HTTP etc.) | BLE (direct GATT) | BLE + gateway | BR/EDR + PAN | BLE + IPSP |
| --- | --- | --- | --- | --- | --- |
| Periodically send a few octets | Yes (heavy fixed overhead) | Best | Best | Extra work | Yes |
| Transfer large files | Best | No (low effective throughput) | Extra work | Yes | Extra work |
| Run for a long time on a coin cell | No | Best (depends on settings) | Best (device side) | Extra work | Yes |
| Device straight to the cloud | Best | No (needs a relay) | Yes | Yes (via NAP) | Yes (via router) |
| Control directly from a browser | Best (HTTP / WebSocket) | Yes (within Web Bluetooth's support) | Best (on the HTTP side) | Extra work (reachable over HTTP if the device's OS supports PAN; not via Web Bluetooth) | Extra work (reachable over HTTP if the device's OS supports IPSP; not via Web Bluetooth) |
| Control from a smartphone app | Yes (assumes the same network) | Best | Yes | Extra work (check OS support) | Extra work (check OS support) |
| Use existing IP tooling | Best | No | Yes (beyond the gateway) | Best | Yes |
| Breadth of device support | Best | Best | Best | Extra work (check per product) | Extra work (check per product) |

Here is an example of how to read it. Suppose you want "readings from a coin-cell sensor shown on a cloud dashboard." The options rated "Best" or "Yes" for both "Run for a long time on a coin cell" and "Device straight to the cloud" are BLE + gateway and BLE + IPSP, and once breadth of device support is factored in, BLE + gateway is the first candidate. PAN and IPSP collect many "Extra work" ratings because support is decided product by product, and not every device supports them. For real-time communication on the browser side, [choosing between WebSocket and SSE](/blog/websocket-sse-realtime-architecture-decision-guide) is also useful.

---

## 17. Adoption patterns in IoT and business systems

These are architectural patterns, not specific case studies.

**Pattern A: connect a smartphone or browser directly to a BLE device.** A smartphone app, or a web page using Web Bluetooth, acts as the GATT client. The Web Bluetooth specification restricts a web page to the **Central role**, so two browsers never connect to each other directly over Bluetooth. In Chrome it is available without flags on Android, ChromeOS, macOS and Windows, but Safari and Firefox do not support it (details in the [Web Bluetooth API implementation guide](/blog/web-bluetooth-api-javascript-typescript-ble-guide)). No internet connection is needed between the phone or computer and the device, but Web Bluetooth is limited to secure contexts, so the page must first be served and loaded over HTTPS, and flashing the firmware is a separate step.

**Pattern B: from a BLE device through a gateway to the cloud.** The sensor runs on a battery for a long time, and the complexity of IP is concentrated in the gateway. The [ESP32 and Web Bluetooth chat app](/blog/web-bluetooth-esp32-ble-chat-app-implementation), in which two browsers use an ESP32 as a relay, is a variant where the ESP32 relays over GATT.

**Pattern C: from the device straight to the cloud over Wi-Fi.** The device joins a LAN over Wi-Fi and sends data to the cloud over HTTPS. My own IoT product, [uomamori](/labs/uomamori), is one example of this pattern: it is designed around an ESP32-C3 with ESP-IDF, sending data over HTTPS through Wi-Fi (the hardware has not shipped yet). The challenge in this pattern is how to get the Wi-Fi SSID and password onto the device; uomamori uses the SoftAP approach, in which the device temporarily becomes an access point itself.

---

## 18. Common misconceptions, corrected from primary sources

| Common claim | Reality | Section |
| --- | --- | --- |
| Bluetooth is OSI layer 7 | It does not fit into one layer; any OSI mapping is a conceptual approximation | 5 |
| Bluetooth cannot do IP / Bluetooth always uses IP | IP can be carried with PAN/BNEP or IPSP, but a GATT-only configuration does not use IP | 7-9 |
| BLE can send at most 20 bytes | 20 is the default ATT_MTU of 23 minus 3. The MTU can be negotiated, and attribute values can be up to 512 octets | 6 |
| Bluetooth FTP uses TCP port 21 | The File Transfer Profile uses OBEX, not TCP | 10 |
| Bluetooth can connect up to 7 devices | 7 is the limit of active Peripherals in a BR/EDR piconet. For LE it depends on the implementation | 3 |
| The latest version is Bluetooth 6.0 | The latest is 7.0 (adopted November 2, 2026) | 4 |

Other misconceptions ("BLE is Classic's low-power mode," "Bluetooth 7 is 7.5 Mbps," "browsers connect to each other directly," and so on) are covered in their respective sections.

---

## 19. Summary

Whether HTTP or TCP works over Bluetooth comes down to whether the configuration carries IP via PAN/BNEP or IPSP. Compare speeds by effective throughput, and judge power by measuring on real hardware.

The implementation side of controlling BLE devices from the browser is covered in the [Web Bluetooth API implementation guide](/blog/web-bluetooth-api-javascript-typescript-ble-guide), and the GATT relay application in the [ESP32 and Web Bluetooth chat app](/blog/web-bluetooth-esp32-ble-chat-app-implementation).
