Course: CS176C — Advanced Topics in Internet Computing, Spring 2026
Instructor: Arpit Gupta, UC Santa Barbara
Type: Supplementary background — for curious students, not required for the midterm
Pre-requisite: L5-L7 (CSMA/CA, DCF, WiFi evolution, OFDMA, overhead analysis)
Why This Note Exists
In L5-L7, we treated the physical layer as a black box that delivers a “PHY rate” and imposes a “162 us preamble overhead.” You now know that higher-order modulation needs higher SNR, that MIMO multiplies spatial streams, and that OFDMA splits subcarriers into Resource Units. This note opens the black box and explains how each of those mechanisms works.
No signal processing background is assumed. We will build every concept from the ground up: what a radio signal physically is, what it means to “modulate” it, what a “symbol” is, and how OFDM combines many signals into one. Every factual claim is cited to a source in the course NotebookLM corpus.
0. What Is a Radio Signal?
Before we can understand WiFi’s physical layer, we need to understand the raw material it works with: electromagnetic waves.
0.1 The Carrier Wave: A Sine Wave You Can Change
A radio signal is an electromagnetic wave that carries information through space without physical connections [1][2]. At its most basic mathematical form, this signal is a sine wave — a smooth, repeating oscillation [3][4].
A sine wave has exactly three properties you can adjust:
- Amplitude: The height or strength of the wave [3][5]. Think of it as the “volume knob” — how tall the peaks are.
- Frequency: The rate at which the wave oscillates, measured in cycles per second (Hertz) [3][5]. A 5 GHz WiFi signal completes 5 billion cycles every second.
- Phase: The starting point of a wave cycle relative to a fixed reference [3][4]. Imagine two identical sine waves — if one starts slightly later than the other, they differ in phase.
The carrier wave is a reference sine wave at a specific, steady frequency that the transmitter will modify to represent data [5][6]. The carrier frequency (for example, 5.2 GHz) defines which radio “lane” the network operates in [7][8]. Different technologies coexist by operating in non-overlapping frequency bands — WiFi uses the 2.4 GHz, 5 GHz, and 6 GHz unlicensed bands, while cellular operators pay billions for exclusive licensed bands [7][9].
0.2 Digital Modulation: Changing the Carrier to Encode Bits
Data is simply bits — 1s and 0s [3][4]. The core idea of wireless communication is this: if you change one of the carrier wave’s three properties (amplitude, frequency, or phase), the receiver can detect that change and recover the original bits [5][6].
Modulation is the “technique adopted in wireless systems to convert digital data (bits) into analog signals by changing the shape of a reference signal, called carrier, within a set of possible modifications” [5][6].
When we restrict those changes to a finite set of discrete states (rather than a continuous range), we call it digital modulation or keying — the signal “shifts” between specific states:
- Phase Shift Keying (PSK): Change the phase of the carrier to represent data [5][6]. For example, a “1” can be a sine wave starting at the origin (going up first), while a “0” is a negative sine wave starting in the opposite direction — a 180-degree phase shift [3][4].
- Frequency Shift Keying (FSK): Switch between discrete frequencies. This was used in early 802.11 FHSS (Frequency Hopped Spread Spectrum) systems [10][11].
- Amplitude Shift Keying (ASK): Switch between discrete amplitude levels. The 802.11ba (Wake-Up Radio) standard uses a form of ASK called On-Off Keying (OOK) to minimize power consumption for IoT devices [12][13].
0.3 What Is a “Symbol”?
A symbol is the fundamental unit of a WiFi transmission [14][15]. It is a modulated electromagnetic wave pattern that represents a specific set of bits [16][17].
Think of it this way: the transmitter holds the carrier wave in one particular state (a specific combination of amplitude and phase) for a fixed duration. That one state, held for that duration, is one symbol. Then it switches to the next state for the next group of bits.
Symbol duration is not arbitrary — it is tied to the physics of the channel. In OFDM systems (which WiFi uses), the symbol duration T equals 1 divided by the subcarrier spacing: T = 1 / delta_F [14][15]. Legacy WiFi (802.11a/g/n/ac) uses a 3.2 us symbol duration, while 802.11ax (WiFi 6) extended this to 12.8 us to improve robustness in outdoor environments [14][15][18][19].
0.4 Bits per Symbol: Why 64-QAM Carries 6 Bits
Here is the key insight that drives WiFi’s throughput evolution. If you have only 2 possible states (like BPSK — two phase positions), each symbol carries 1 bit. But if you can reliably distinguish 4 states, each symbol carries 2 bits (since log2(4) = 2). With 64 states, each symbol carries 6 bits (since log2(64) = 6) [5][6].
The number of bits per symbol equals log2(M), where M is the number of distinct states (constellation points) in the modulation scheme [5][6][3][4]. This logarithmic relationship — bits = log2(M) — explains both the power of higher-order modulation and its diminishing returns [20][21].
The progression through WiFi generations:
| Modulation | States (M) | Bits per Symbol | WiFi Generation |
|---|---|---|---|
| BPSK | 2 | 1 | 802.11a/g (MCS 0) |
| QPSK | 4 | 2 | 802.11a/g (MCS 1-2) |
| 16-QAM | 16 | 4 | 802.11a/g (MCS 3-4) |
| 64-QAM | 64 | 6 | 802.11a/g (MCS 5-7) |
| 256-QAM | 256 | 8 | 802.11ac / WiFi 5 [22][23] |
| 1024-QAM | 1024 | 10 | 802.11ax / WiFi 6 [22][23] |
| 4096-QAM | 4096 | 12 | 802.11be / WiFi 7 [22][23] |
Going from 1024-QAM to 4096-QAM adds only 2 bits — a 20% rate increase — while requiring dramatically higher signal quality [24][25].
1. Modulation: Encoding Bits onto the Carrier
1.1 From Phase Shifting to QAM
Let us build up from the simplest case.
BPSK (Binary Phase Shift Keying): The simplest digital modulation. A “1” can be represented as a sine wave that begins at the origin (goes up, then down, then up), and a “0” can be a negative sine wave (down, up, down) [3][4]. These are two phase states — 0 degrees and 180 degrees — so each symbol carries exactly 1 bit.
QPSK (Quadrature Phase Shift Keying): Take the same sine wave and shift its starting point by pi/2 (90 degrees). Now you have 4 distinct phase positions, and each represents a unique 2-bit pair (00, 01, 10, 11) [3][4].
Higher-order PSK: Shifting by pi/4 gives 8 positions (3 bits). Shifting by pi/8 gives 16 positions (4 bits). But at some point, phase-only changes become too similar for the receiver to distinguish [3][4].
QAM (Quadrature Amplitude Modulation): The breakthrough — vary both amplitude and phase simultaneously [3][4][5][6]. “Alterations can be made to the amplitude of the wave as well, creating more opportunities for modulation and multiplying bitrate. This can be represented by a 2D map of amplitude and phase alterations and is known as constellation mapping” [3][4].
1.2 The Constellation Diagram
A constellation diagram is a 2D map in the complex plane (I = In-phase axis, Q = Quadrature axis). Each point represents a unique combination of amplitude and phase changes applied to the carrier wave [5][6]. The stream of coded bits is divided into groups, and each group is mapped into a constellation point, generating a modulated symbol [5][6].
For example:
- 16-QAM places 16 points in a 4x4 grid — each encodes 4 bits
- 64-QAM uses an 8x8 grid — 6 bits per symbol
- 256-QAM uses a 16x16 grid — 8 bits per symbol [5][6]
- 4096-QAM uses a 64x64 grid — 12 bits per symbol, introduced in WiFi 7 [24][25]
The receiver’s job: measure the received signal’s amplitude and phase, determine which constellation point was most likely sent, and extract the corresponding bits [5][6].
1.3 Why Higher-Order Modulation Requires Higher SNR
Here is the fundamental tradeoff. As you pack more constellation points into the same total signal power, the distance between neighboring points shrinks [20][21].
The corpus uses a vivid analogy: think of it as packing puzzle pieces in a box. “By shrinking the height of the puzzle pieces, we can fit more pieces in each box” — that is higher-order QAM. But “thinner puzzle pieces do not have this luxury; even slight disturbances may destroy the puzzle piece beyond recognition” [3][4][20][21].
Noise causes the received signal to “jitter” around the true constellation point. If the jitter exceeds half the distance to the nearest neighbor, the receiver picks the wrong point — a bit error [20][21]. The signal must be “consistently clean enough (interference-free) to allow the receiver to differentiate slight differences in signal” [20][21].
Approximate SNR requirements:
| Modulation | Approximate Minimum SNR | Notes |
|---|---|---|
| BPSK | ~6 dB | Most robust [5][6] |
| QPSK | ~9 dB | [5][6] |
| 16-QAM | ~15 dB | [5][6] |
| 64-QAM | ~21 dB | [5][6] |
| 256-QAM | ~25 dB | “Pristine RF environments” needed [26] |
| 1024-QAM | ~35 dB | “Very high SNR thresholds… close proximity between AP and client required” [26][27] |
| 4096-QAM | ~40-42 dB | “Clients have to be very close to the Access Point, within a few feet” [28] |
1.4 BER vs. SNR: The Waterfall Curves
Each modulation scheme produces a characteristic Bit Error Rate (BER) vs. SNR curve [29]:
- Shape: a “waterfall” — BER is essentially random at very low SNR, then plunges exponentially once a threshold is crossed [30][31].
- Ordering: BPSK reaches low BER at the lowest SNR. Each step up to higher QAM shifts the waterfall curve rightward by several dB [29][32].
- The coding rate effect: FEC coding (rate 1/2 vs. 5/6) shifts the curve leftward. A 1/2 coding rate means that “for each information bit, there is another bit for FEC purposes” — more redundancy improves reliability at the same SNR, but carries fewer data bits [5][6].
What this means in practice: at a given SNR, there is one MCS that maximizes throughput. The rate adaptation algorithm hunts for this sweet spot per-packet [5][6].
2. OFDM: Many Subcarriers Playing Together
2.1 The Problem: Frequency-Selective Fading
OFDM “effectively mitigates the frequency-selective fading responsible for ISI, i.e., the phenomenon for which delayed copies of a transmitted signal carrying previous symbols overlap at the receiver with the signal carrying the current symbol” [14][15].
In a 20 MHz wideband channel, multipath reflections cause some frequencies to experience deep fades (near-zero signal strength) while adjacent frequencies are fine. A single fast serial stream across the whole band would be devastated by this.
2.2 The Solution: Many Narrow Subcarriers in Parallel
OFDM divides the available bandwidth into many partially overlapping but orthogonal subcarriers [33]. Each subcarrier is an independent signal at a specific frequency [33][34]. When a device transmits using OFDM, “it transmits data on the subcarriers in parallel, using multiple smaller sub-signals instead of one large signal over the entire channel, improving signal efficiency and resilience” [35][36].
The orchestra analogy: Think of OFDM like an orchestra where each instrument plays at its own frequency. The violins play high notes, the cellos play low notes, and the basses play the lowest — all simultaneously. Each instrument (subcarrier) carries its own independent melody (data), and together they compose a rich, full sound (the composite waveform). The conductor (the AP) orchestrates when each instrument plays and what it plays [33][37].
Because each subcarrier is very narrow (312.5 kHz in legacy WiFi, or 78.125 kHz in WiFi 6), it experiences approximately “flat” fading — the channel attenuation is roughly constant across its bandwidth [33][38]. A deep fade might destroy one subcarrier, but its neighbors survive.
There are three types of subcarriers [35][36]:
- Data subcarriers: used to transmit data
- Pilot subcarriers: used for synchronization between sender and receiver
- Guard subcarriers: used to avoid interference at band edges
Subcarrier counts per generation:
| Parameter | 802.11a/g/n/ac (Legacy) | 802.11ax (WiFi 6) and beyond |
|---|---|---|
| Total subcarriers (20 MHz) | 64 | 256 |
| Data subcarriers | 52 | 234 |
| Pilot subcarriers | 4 | 8+ |
| Subcarrier spacing | 312.5 kHz | 78.125 kHz |
| Symbol duration (data only) | 3.2 us | 12.8 us |
| Guard interval options | 0.4 or 0.8 us | 0.8, 1.6, or 3.2 us |
Sources: [38][39] for legacy 64/52 subcarriers at 312.5 kHz; [38][39] for 802.11ax 256/234 at 78.125 kHz; [18][19] for the 4x relationship.
Why 802.11ax quadrupled the subcarrier count: not primarily for more data capacity per user, but to enable OFDMA — subdividing the channel into Resource Units (RUs) that can be assigned to different users simultaneously. The smallest RU (26 subcarriers, ~2 MHz) serves one user; a 20 MHz channel supports up to 9 simultaneous users [40][41][42].
2.3 IFFT and FFT: How the Magic Works
The IFFT/FFT pair is what makes OFDM computationally practical. This breakthrough was described by Weinstein and Ebert in 1971 [43].
At the transmitter (IFFT): You have N_ST constellation points — one data symbol per subcarrier — sitting in the frequency domain. “The signal is usually obtained by forwarding the N_ST modulated symbols — referred to as OFDM samples — through an IFFT block. The set of N_ST samples at the output of the IFFT block is referred to as an OFDM symbol” [14][15].
Mathematically, the IFFT performs a summation of independent sine waves, each at its own subcarrier frequency and with its own amplitude and phase set by the constellation point [33][34]. The formula is:
x_m(t) = sum over k of a_{m,k} * e^{j*2*pi*(f_c + k*delta_F)*t}
where a_{m,k} is the constellation point on the k-th subcarrier [33][34]. This single operation replaces what would otherwise require N_ST separate analog up-converters — one per subcarrier.
At the receiver (FFT): The receiver samples the incoming waveform, runs an FFT, and recovers the N_ST constellation points — one per subcarrier — for demodulation [14][15]. The FFT decomposes the single incoming wave back into its constituent frequency components, like a digital prism splitting white light into colors.
This is computationally efficient: an N-point FFT costs O(N log N) operations [43].
2.4 The Cyclic Prefix: Handling Multipath Echoes
Multipath reflections cause delayed copies of symbol N to arrive while the receiver is decoding symbol N+1. This is Inter-Symbol Interference (ISI) [14][15].
The fix: “A cyclic prefix (CP) is added at the beginning of each OFDM symbol by repeating the last portion of the same symbol. This acts as a GI (guard interval) between symbols, helping to reduce ISI even in environments with significant delay spread” [14][15].
Why it works: The CP creates a sacrificial buffer. If the maximum multipath delay is shorter than the CP duration, all delayed copies of the previous symbol land within the CP of the current symbol. The receiver discards the CP and processes only the clean portion.
CP duration by generation:
| CP Option | Duration | Environment |
|---|---|---|
| Legacy (default) | 0.8 us | Indoor, minimal delay spread [44][45] |
| 802.11ax option 1 | 1.6 us | Moderate outdoor, UL MU-MIMO/OFDMA [44][45] |
| 802.11ax option 2 | 3.2 us | Extreme outdoor delay spread [44][45] |
The overhead cost: CP is “wasted” time — no new data is transmitted. At 0.8 us CP with 3.2 us symbol (legacy), CP overhead = 0.8/4.0 = 20%. At 0.8 us CP with 12.8 us symbol (802.11ax), CP overhead = 0.8/13.6 = 5.9%. “Longer symbol duration reduces cyclic prefix overhead relative to the symbol length, increasing efficiency, especially indoors” [14][15].
3. The Complete TX/RX Chain
3.1 Transmitter Chain (Bits to RF)
The PHY receives data from the MAC in a PHY Service Data Unit (PSDU), selects a Modulation and Coding Scheme (MCS), prepends a preamble, processes everything into an RF signal, and radiates it through the antennas [46][47].
| Step | What Happens | Why |
|---|---|---|
| 1. Scrambling | XOR bits with a pseudo-random scrambling sequence | “Reduces the probability of long sequences of bits equal to zeros or ones, which helps time synchronization at the receiver” [48][49] |
| 2. FEC Encoding | Add redundancy via Binary Convolutional Code (BCC) or Low-Density Parity Check (LDPC) | Enables Forward Error Correction — the receiver can correct bit errors without retransmission. Coding rate (1/2, 2/3, 3/4, 5/6) sets the redundancy level [48][49][5][6] |
| 3. Interleaving / Tone Mapping | For BCC: two-step permutation of coded bits. For LDPC: permutation of modulated symbols across separated subcarriers | “Avoid long sequences of noisy bits on the BCC decoder and thus improve transmission robustness to burst errors” [50][51] |
| 4. Constellation Mapping | Map groups of N_BPSC bits to constellation points in the I/Q plane | “The stream of coded bits is divided into groups of N_BPSC bits, each of which is mapped into a constellation point generating a modulated symbol” [5][6] |
| 5. Pilot Insertion | Insert known OFDM symbols (pilots) at specific subcarrier positions | “For receiver-transmitter synchronization purposes to make the data detection robust against frequency offsets and phase noise” [52][53] |
| 6. Spatial Stream Mapping | Distribute modulated symbols across N_SS spatial streams | Enables MIMO. A beamforming steering matrix may be applied if channel state is known [52][53] |
| 7. OFDM via IFFT | Pass N_ST frequency-domain symbols through an Inverse Fast Fourier Transform | Converts frequency-domain constellation points into a time-domain OFDM symbol [14][15] |
| 8. Cyclic Prefix | Copy the last portion of the OFDM symbol and prepend it | “Acts as a GI between symbols, helping to reduce ISI even in environments with significant delay spread” [14][15] |
| 9. D/A Conversion | Convert discrete digital samples to a continuous analog waveform | Null subcarriers near the center frequency provide protection from “transmit center frequency leakage, DAC and ADC offsets” [33] |
| 10. RF Up-conversion | Shift the baseband signal to the carrier frequency (e.g., 5.2 GHz) | “The RF signals to be transmitted over the different antennas are obtained through a frequency up-conversion at the specific operating frequency defined by the selected Wi-Fi channel” [54][55] |
3.2 Receiver Chain (RF to Bits)
The receiver reverses the pipeline:
| Step | What Happens |
|---|---|
| 1. RF Down-conversion | Shift incoming RF signal back to baseband |
| 2. Preamble Detection + Sync | Use the L-STF for signal detection and AGC; use the L-LTF for fine timing/frequency recovery and channel estimation [56][57] |
| 3. FFT | Convert time-domain samples back to frequency-domain symbols — one per subcarrier |
| 4. Equalization | Apply a decoding matrix using channel estimates from the LTF to undo frequency-selective fading and separate spatial streams [58][59] |
| 5. Demodulation | Detect which constellation point each received symbol is closest to; extract the bit group |
| 6. Deinterleaving | Reverse the interleaver permutation (BCC) or tone mapper (LDPC) |
| 7. FEC Decoding | Use the redundancy to correct remaining errors |
| 8. Descrambling | XOR with the same scrambling sequence to recover original data bits |
3.3 Where the Preamble Fits
The preamble is transmitted before the data and is always sent at the lowest, most robust settings so every device in earshot can decode it [56][57].
Preamble fields (legacy portion, present in every frame):
| Field | Duration | Purpose |
|---|---|---|
| L-STF (Legacy Short Training Field) | 8 us | Signal detection, AGC, coarse frequency recovery [56][57] |
| L-LTF (Legacy Long Training Field) | 8 us | Fine timing recovery, channel estimation — the receiver builds its model of the channel [56][57] |
| L-SIG (Legacy Signal Field) | 4 us | Frame rate and length metadata; tells legacy devices how long the medium will be busy [56][57] |
Total legacy preamble: 20 us. This duration is confirmed across multiple corpus sources [60][61]. It is the irreducible “overhead tax” we discussed in L7 — it takes 20 us regardless of whether the data portion is sent at 6 Mbps or 9.6 Gbps [56][57].
Why the preamble does not scale with data rate: Unlike the data payload, the preamble is always transmitted at the most robust, lowest mandatory rate. This is a backward compatibility requirement — a 1999-era 802.11a device hearing a 2024-era 802.11be frame must be able to decode the L-SIG to understand how long the medium will be busy, so it can set its Network Allocation Vector (NAV) and avoid collisions [62][63].
Modern formats (HE, EHT) add additional fields after L-SIG for MIMO/OFDMA parameters. 802.11be introduced a Universal SIG (U-SIG) field spanning two OFDM symbols for version detection [64]. Total preamble ranges from roughly 44 us (single-user) to 164 us (multi-user) depending on the number of spatial streams and users [46][47].
4. MIMO: Exploiting Space
4.1 Spatial Diversity vs. Spatial Multiplexing
MIMO uses multiple antennas (spaced at least half a wavelength apart to ensure independent fading) for two fundamentally different purposes [65][66]:
| Spatial Diversity | Spatial Multiplexing | |
|---|---|---|
| Goal | Improve reliability | Improve throughput |
| Method | Send the same data on multiple antenna paths; combine at receiver | Send different data streams on different antenna paths simultaneously |
| When useful | Low SNR, long range, high interference | High SNR, rich scattering (many multipath reflections) |
| Capacity scaling | No increase in peak rate; reduces fading drops | Linear: rate scales with min(N_TX, N_RX) |
| WiFi example | 802.11n STBC (Space-Time Block Code) mode | 802.11n 4x4 spatial multiplexing (4 independent streams) |
Sources: [65] for the diversity-multiplexing tradeoff (Zheng and Tse, 2003); [66] for WiFi MIMO overview; [67][68] for the half-wavelength spacing and independent fading.
The tradeoff is fundamental (Zheng and Tse, 2003): you cannot simultaneously maximize both diversity and multiplexing gain. The system (or rate adaptation algorithm) must choose based on current channel conditions [65].
4.2 Beamforming: Steering Without Moving
Beamforming adjusts the phase and amplitude of the signal at each antenna element so that the signals constructively interfere at the intended receiver and destructively interfere elsewhere [66][68].
How it works physically:
- Consider a 4-antenna AP transmitting to a single client.
- Each antenna transmits the same data, but with a carefully computed phase offset.
- At the client’s location, the 4 wavefronts arrive and add up constructively (in phase). At other locations, they partially cancel.
- The result: a “beam” of focused RF energy aimed at the client, increasing its received SNR [66][68].
The steering matrix: the AP applies a beamforming steering matrix to the modulated symbols before transmission. This matrix is derived from knowledge of the channel — obtained via sounding. Common algorithms: Zero-Forcing (ZF) and MMSE [58][59].
4.3 Channel Sounding: How the AP Learns the Channel
The AP needs to know the Channel Frequency Response (CFR) matrix H — a complex-valued matrix of dimension N_RX x N_TX for each OFDM subcarrier — to compute the steering matrix [69][70].
802.11 uses explicit sounding (the only standardized mechanism) [69]:
| Step | What Happens |
|---|---|
| 1. NDPA | AP sends a Null Data Packet Announcement — “sounding is starting” |
| 2. NDP | AP sends a Null Data Packet containing Long Training Fields (LTFs), one per spatial stream [69][70] |
| 3. Channel Estimation | The STA uses the known LTF patterns to estimate H for every subcarrier [69][70] |
| 4. SVD Compression | The STA compresses H via Singular Value Decomposition — feeds back only rotation angles (phi, psi), not the raw matrix [71] |
| 5. Feedback | The STA sends a compressed beamforming feedback frame (unencrypted, for speed) [71] |
| 6. Steering Matrix | The AP reconstructs V_k matrices from the angles and computes the steering matrix [72] |
Sounding rate: suggested every 10 ms to track channel changes [69]. This is a significant overhead cost.
4.4 MU-MIMO Constraints
MU-MIMO (Multi-User MIMO, introduced in 802.11ac for downlink, extended to uplink in 802.11ax) allows the AP to serve multiple clients simultaneously on different spatial streams [73][74].
Why clients must be spatially separated:
- The AP forms radiation nulls to prevent one client’s stream from interfering with another’s [75][76].
- If clients are in the same direction relative to the AP, their channel signatures are correlated — the AP cannot form distinct nulls, causing inter-user interference [75][77].
- High correlation forces the system to drop to lower MCS, negating the throughput benefit [75].
The sounding overhead problem:
- MU-MIMO requires sounding each client before transmission.
- For a 320 MHz, 16-antenna system, compressed feedback alone is ~22.4 kB per client, consuming ~7.5 ms of airtime at typical feedback rates [78].
- With a 10 ms sounding interval, that is 75% protocol overhead — leaving only 25% for actual data [78].
- This is why MU-MIMO is often less efficient than SU-MIMO in mobile or high-density indoor environments. Some enterprise APs disable MU-MIMO entirely [75][77].
5. Rate Adaptation: The Closed Loop
5.1 Who Decides
The transmitter selects the MCS for each data frame [5][6]. The standard does not mandate a specific rate adaptation algorithm — it is implementation-dependent [5][6].
5.2 What Signals Trigger Rate Changes
| Signal | How It Works |
|---|---|
| RSSI / SNR | Direct or indirect measurement of link quality. Higher SNR permits higher MCS [5][6] |
| Frame loss (ACK timeout) | If the expected ACK is not received, the frame is retransmitted. High loss rates (varying 10%-80% on production APs, ~30% average) signal the need to downshift MCS [79][80] |
| CQI feedback (802.11ax+) | The receiver sends Channel Quality Indicator values (SVD singular values), enabling precise MCS selection at the transmitter [58][59] |
5.3 How Fast
Rate adaptation operates per-packet or per-frame-burst — the MCS can change for every transmission opportunity. The algorithm “tests” higher rates when conditions appear good and falls back when frame errors increase [5][6][79].
5.4 The Rate Adaptation Principle
Low MCS values (low modulation order, high coding redundancy) “lead to more robust transmissions at the expense of reduced throughput and should be preferred in low SNR conditions.” High MCSs “require increasingly higher SNR to maintain robustness against noise and interference but increase data rates by modulating more bits together per symbol and reducing coding redundancy” [5][6].
The rate adaptation principle: maximize effective throughput, not PHY rate. The best MCS is the one where PHY_rate x (1 - FER) is maximized [5][6].
6. How the PHY Constrains the MAC
This section connects the PHY mechanisms above to the MAC-layer behaviors you studied in L5-L7.
6.1 The Preamble Is the Overhead Tax
Every frame — data, ACK, RTS, CTS — begins with a preamble:
| Component | Duration | Rate | Notes |
|---|---|---|---|
| Legacy preamble (L-STF + L-LTF + L-SIG) | 20 us | Fixed (BPSK, rate 1/2) | Cannot be sped up — must be decodable by all devices [56][57] |
| HE/EHT preamble extensions | 24-144 us additional | Fixed per spec | Scales with N_SS and user count [46][47] |
| Typical 802.11ax total preamble | ~44 us (SU) to ~164 us (MU) | – | – |
This is why the overhead dominates at high data rates. At 9.6 Gbps PHY rate, a 1500-byte frame takes only ~1.25 us to transmit — but the preamble still takes ~44 us. The data is ~2.8% of the total frame time.
6.2 Why ACK Is Sent at Basic Rate
[NOT IN CORPUS — needs external source. The corpus does not contain an explicit explanation of why ACK uses the basic rate. Standard textbook reasoning: the ACK must be reliably received, so it uses the lowest mandatory rate to maximize reception probability.]
6.3 Why SIFS and DIFS Are Fixed by Radio Turnaround
[NOT IN CORPUS — needs external source. The corpus confirms the SIFS and DIFS values (SIFS = 16 us, DIFS = 34 us, slot time = 9-10 us per [60][61]) but does not explicitly explain that SIFS is bounded by radio turnaround time. Standard textbook reasoning: SIFS accounts for the hardware switching time from receive to transmit mode.]
6.4 Summary: Where PHY Meets MAC
| PHY Mechanism | MAC Consequence |
|---|---|
| Preamble always at lowest rate | Fixed overhead per frame, dominates at high PHY rates |
| Higher MCS needs higher SNR | Rate adaptation determines effective throughput, not just PHY rate |
| OFDMA subcarrier granularity | AP can schedule multiple users per TXOP (WiFi 6+) |
| MIMO sounding overhead | MU-MIMO may lose to SU-MIMO when sounding cost exceeds multiplexing gain |
| Radio turnaround time | SIFS/DIFS are physics-bounded, not protocol choices |
Bibliography
NotebookLM Corpus Sources (103-source medium access collection):
[1] Singh, U., “Demystifying Wi-Fi: A Comprehensive Comparison of Wi-Fi 5, 6, and 7” (2023). Source ID: bc0ba17e. — Wi-Fi overview covering radio wave fundamentals.
[2] Singh, U., “Demystifying Wi-Fi” (alternate rendering). Source ID: ce245832. — Same source, alternate format.
[3] WiFi evolution student survey paper. Source ID: fb93dd5c. — Modulation explained via sine wave cycle representations, phase shifting, constellation mapping, puzzle piece analogy.
[4] WiFi evolution student survey paper (alternate). Source ID: a0a29620. — Same content.
[5] Bellalta, B. et al., “Wi-Fi: Twenty-Five Years and Counting,” IEEE Communications Surveys & Tutorials (2025). Source ID: a82adcb6. — Comprehensive WiFi PHY/MAC tutorial covering the full signal processing chain, MCS definition, constellation points, modulation order, PSK and QAM.
[6] Bellalta, B. et al., “Wi-Fi: Twenty-Five Years and Counting” (alternate rendering). Source ID: ff86991a. — Same source, alternate format.
[7] Coleman, D., “Wi-Fi 6 (802.11ax) Design Concepts.” Source ID: dc8e8b7e. — Spectrum allocation, licensed vs. unlicensed bands, channel numbering.
[8] 802.11 Wikipedia article. Source ID: 1bb69782. — IEEE 802.11 frequency bands (2.4, 5, 6, 60 GHz).
[9] [7, licensed vs. unlicensed bands section] — Cellular operators pay billions for exclusive licensed spectrum; unlicensed bands are free but subject to interference.
[10] Daneshgaran, F. et al., “Saturation throughput analysis of IEEE 802.11 in the presence of non ideal transmission channel.” Source ID: 5f0e6c2b. — 802.11b PHY: FHSS uses FSK, DSSS uses DPSK/CCK.
[11] [10, same source alternate rendering]. Source ID: 04f580bc.
[12] 802.11 Wikipedia standards comparison table. Source ID: 36d52651. — 802.11ba Wake-Up Radio uses multi-carrier OOK.
[13] 802.11 Wikipedia standards comparison table (alternate). Source ID: 6d136ae6. — Same content.
[14] [5, Section III-C, OFDM and IFFT] — IFFT converts frequency-domain symbols to time-domain OFDM symbol; cyclic prefix added; symbol duration T = 1/delta_F.
[15] [6, Section III-C, OFDM and IFFT] — Same content.
[16] Meraki documentation. Source ID: 79797f6a. — “Each stream of data transmitted on the subcarriers is made up of a series of modulated wave patterns called symbols.”
[17] Meraki documentation (alternate). Source ID: c9cbe44e. — Same content on symbols and symbol time.
[18] Semfionet (industry blog). Source ID: 3527427e. — 802.11ax uses 4x subcarriers vs. 802.11ac; OFDM symbols 4x longer; symbol duration 12.8 us.
[19] Semfionet (alternate). Source ID: bb2fa888. — Same content.
[20] [3, Section 2.7.2, Limitations of modulation] — Puzzle piece analogy; logarithmic diminishing returns; interference forces MCS reduction.
[21] [4, Section 2.7.2, Limitations] — Same content.
[22] [5, Table I, WiFi generation comparison] — MCS levels, modulation orders, and peak rates per standard (802.11b through 802.11bn).
[23] [6, Table I] — Same content.
[24] WiFi 7 4096-QAM overview. Source ID: b102d89e. — “4096-QAM achieves a 20% rate increase over 1024-QAM.”
[25] WiFi 7 4096-QAM overview (alternate). Source ID: f2ddb0b5. — Same content.
[26] Coleman, D., “Wi-Fi 6 Design Concepts.” Source ID: e7f138d9. — 256-QAM requires ~25 dB SNR; 1024-QAM needs ~35 dB; pristine RF required.
[27] [26, same source] — 1024-QAM SNR threshold confirmation.
[28] WiFi 7 overview (industry source). Source ID: e3107f6f. — 4096-QAM requires ~42 dB SNR; clients must be within feet of AP.
[29] Daneshgaran, F. et al., “Saturation throughput analysis.” Source ID: 5f0e6c2b. — BER formulas for DBPSK, DQPSK, CCK; FER as function of SNR.
[30] PHY-MAC simulation paper. Source ID: ec79913b. — PER vs. SNR waterfall curves.
[31] [30, same source] — PHY-MAC throughput vs. SNR for different MCS settings.
[32] [29, Section on rate adaptation and BER curves] — BER vs. SNR curves for each modulation format.
[33] [5, Section III-C, OFDM] — N_ST partially overlapping orthogonal subcarriers; three types (data, pilot, guard/null); DAC/ADC offset protection via null subcarriers.
[34] [5, Equation 1] — OFDM symbol as summation of modulated subcarriers.
[35] Meraki documentation. Source ID: c9cbe44e. — “transmits data on the subcarriers in parallel, using multiple smaller sub-signals instead of one large signal.”
[36] Meraki documentation. Source ID: 79797f6a. — Three types of subcarriers; OFDM 64 at 312.5 kHz, OFDMA 256 at 78.125 kHz.
[37] [5, Section V, OFDMA and AP orchestration] — AP orchestrates uplink/downlink transmissions via trigger frames.
[38] [5, Figure 4] — 802.11ac: 52 data subcarriers at 312.5 kHz, 3.2 us symbol. 802.11ax: 234 data subcarriers at 78.125 kHz, 12.8 us symbol.
[39] [6, Figure 4] — Same content.
[40] Ph.D. dissertation on 802.11ax OFDMA. Source ID: 0adfb495. — 26-subcarrier RU = smallest unit; 9/18/37/74 max users for 20/40/80/160 MHz.
[41] Coleman, D., “Wi-Fi 6 Design Concepts.” Source ID: e7f138d9. — 26/52/106/242 subcarrier RU sizes; AP coordinates via trigger frames.
[42] WiFi 6 technology overview. Source ID: d4b41eaa. — OFDMA borrows from 4G LTE; assigns subcarrier sets to individual users.
[43] Literature survey source. Source ID: 3f29e336. — Weinstein and Ebert, 1971: DFT/FFT for OFDM modulation — the breakthrough making OFDM computationally practical.
[44] [5, cyclic prefix options] — 0.8 us (indoor default), 1.6 us (outdoor/UL MU), 3.2 us (extreme outdoor).
[45] [6, cyclic prefix options] — Same content.
[46] [5, Section III-F, PPDU formats] — PPDU structure: preamble + data payload; evolved across WiFi generations for backward compatibility.
[47] [5, Section III-F2, U-SIG in 802.11be] — Universal SIG field spanning two OFDM symbols for version detection.
[48] [5, Section III-A, Physical Layer Block Diagram] — Scrambling via additive scrambler (XOR); FEC via BCC or LDPC.
[49] [6, Section III-A] — Same content.
[50] [5, BCC interleaver section] — Two-step permutation of coded bits to prevent burst errors on the BCC decoder.
[51] [6, LDPC tone mapper section] — Tone mapping for LDPC to ensure consecutive symbols are separated.
[52] [5, Section III-A, pilot insertion and spatial stream mapping] — Pilots for synchronization; spatial streams via round-robin; CSD for backward compatibility.
[53] [6, Section III-A] — Same content.
[54] [5, Section III-A, RF up-conversion] — Baseband signal up-converted to carrier frequency f_c.
[55] [6, Section III-A, RF up-conversion] — Same content.
[56] [5, Section III-F2, Preamble Design] — L-STF for detection/AGC, L-LTF for fine recovery/channel estimation, L-SIG for metadata.
[57] [6, Section III-F2, Preamble Design] — Same content.
[58] [5, Section III-D, MMSE equalization and ZF precoding] — Beamforming steering matrix derivation via ZF and MMSE.
[59] [6, Section III-D, MMSE equalization] — Same content.
[60] MLO delay analysis paper. Source ID: 4a35e25d. — Legacy preamble = 20 us; HE-SU preamble = 52 us; SIFS = 16 us; DIFS = 30 us.
[61] PHY-MAC simulation paper. Source ID: ec79913b. — PHY preamble & header = 20 us; SIFS = 16 us; DIFS = 34 us; slot time = 9 us.
[62] [5, Section IV-A, DCF and NAV] — Virtual carrier sensing: decode PHY header to read Duration/ID field; update NAV to avoid collisions.
[63] [6, Section IV-A, DCF and NAV] — Same content.
[64] [5, Figure 7, U-SIG] — U-SIG spans two OFDM symbols; contains version-independent and version-dependent bits for format detection.
[65] Zheng, L. and Tse, D. N. C., 2003, “Diversity and Multiplexing: A Fundamental Tradeoff in Multiple-Antenna Channels,” IEEE Trans. Information Theory, 49(5). — Via literature survey source (Source ID: c297b045).
[66] [5, Section III-D, Multi-Antenna Systems] — Diversity gain, beamforming gain, multiplexing gain overview; half-wavelength spacing.
[67] Foschini, G. J., 1996, “Layered Space-Time Architecture,” Bell Labs Technical Journal, 1(2). — Linear capacity scaling with min(N_t, N_r). Via Source ID: c297b045.
[68] [5, Section III-D, beamforming] — TxBF precoding, steering matrix, SU-MIMO identification.
[69] [5, Figure 5 and Section III-D, explicit sounding] — NDPA, NDP, LTF-based channel estimation, sounding at 10 ms intervals.
[70] [6, explicit sounding procedure] — Same content.
[71] [5, SVD compression of CFR feedback] — Right unitary matrices, phi/psi angles, quantization, grouping factor N_g.
[72] [5, steering matrix reconstruction from feedback] — AP retrieves angles, reconstructs V_k, computes steering matrix.
[73] WiFi evolution survey (student paper). Source ID: a0a29620. — SU-MIMO in WiFi 4; MU-MIMO in WiFi 5; up to 8 streams in WiFi 6.
[74] Meraki documentation. Source ID: 79797f6a. — 802.11ac DL MU-MIMO (4 clients); 802.11ax UL/DL MU-MIMO (8 clients).
[75] [4, MU-MIMO limitations] — Spatial separation required; same-direction clients cause interference; real-world results vary, sometimes -58%.
[76] [26, MU-MIMO enterprise deployment] — MU-MIMO requires spatial diversity; sounding frames add excessive overhead.
[77] [3, MU-MIMO limitations] — APs in crowded environments may disable MU-MIMO entirely.
[78] [5, sounding overhead analysis] — 16 antennas, 320 MHz: ~22.4 kB feedback, ~7.5 ms airtime per 10 ms sounding interval = 75% overhead.
[79] 802.11 Wikipedia article on frame loss and rate control. Source ID: 36d52651. — Rate control tests different speeds; production AP loss rates 10%-80%, ~30% average.
[80] 802.11 Wikipedia article (alternate). Source ID: 1bb69782. — Frame loss built into 802.11 operation.
Last updated: April 16, 2026. Prepared as supplementary material for CS176C, Spring 2026.