Imported from the-vibey-project/vibey-skills (plugins/radio-technology-for-software-devs/skills/radio-spread-spectrum-ofdm-access-and-sdr/SKILL.md). Install upstream with npx skills add the-vibey-project/vibey-skills --skill radio-spread-spectrum-ofdm-access-and-sdr. Copyright stays with the author.

Radio Technology: Spread Spectrum, OFDM, Coding, Multiple Access, and Software-Defined Radio

Part 3 of 6 of the Radio Technology for Software Developers reference (plugin radio-technology-for-software-devs), covering §10–§16. Sibling skills: radio-intuitions-spectrum-link-budget-and-tradeoffs (§0–§4), radio-antennas-propagation-noise-and-modulation (§5–§9), radio-protocol-stacks-wifi-ble-lpwan-and-gnss (§17–§22), radio-regulatory-security-and-debugging (§23–§26), radio-reference (§27–§32). Section numbers are shared across the set; a reference written as §N → skill points into that sibling skill.

Currency: RF physics and DSP are permanent. Two areas moved. See §24 → radio-regulatory-security-and-debugging for Wi-Fi 7 and 8 and the 6 GHz regulatory status, and the IoT connectivity landscape after the 2G and 3G sunset.

⚠️ Scope. Written for people who write software and now have to make a radio work. Complements an embedded/IoT reference (the devices), an electrical engineering reference (the circuits), and a networking reference (layer 3 and up). This is layers 0–2 and the physics underneath them.

⚠️ GOTCHA boxes mark things that cause intermittent, unreproducible field failures — the expensive kind.

The three ideas that organize this document:

1. ⚠️ RF is where software abstractions leak worst. Every layer above assumes a link that mostly works; RF is a shared, non-deterministic, physically-constrained medium where the same code works on your desk and fails in the field (§1 → radio-intuitions-spectrum-link-budget-and-tradeoffs).
2. ⚠️ The link budget is the single most useful tool you can learn (§3 → radio-intuitions-spectrum-link-budget-and-tradeoffs). Most wireless problems are budget problems, and most of them are diagnosable on paper before you build anything.
3. ⚠️ Choosing the radio is an architecture decision, not a component decision. Range, data rate, power and cost trade against each other in ways physics fixes (§4 → radio-intuitions-spectrum-link-budget-and-tradeoffs, §19 → radio-protocol-stacks-wifi-ble-lpwan-and-gnss), and you cannot fix a bad choice in firmware.

§10. Spread Spectrum

⚠️ Deliberately using much more bandwidth than the data requires, in exchange for robustness.

DSSS   ⚠️ multiply data by a fast pseudo-random code. Spreads energy;
       the receiver correlates to recover it, gaining PROCESSING GAIN.
       Interference and multipath are suppressed. 802.15.4, GPS, older Wi-Fi
FHSS   ⚠️ hop the carrier around a channel set. If one channel is jammed
       you lose one hop, not the link. ⚠️ Bluetooth — 1600 hops/second
CSS    ⚠️ chirp spread spectrum — LoRa. Frequency sweeps across the band
CDMA   multiple users, different codes, same band and time

⚠️ LoRa's spreading factor is the clearest illustration of §4 → radio-intuitions-spectrum-link-budget-and-tradeoffs in a single knob: ⚠️ SF7 → SF12 roughly doubles airtime per step, adds a few dB of sensitivity per step, and cuts the data rate. SF12 reaches furthest and can occupy the channel for seconds per message — ⚠️ which collides directly with duty-cycle regulation (§23 → radio-regulatory-security-and-debugging) and destroys network capacity if used carelessly. Adaptive Data Rate exists to push devices to the lowest SF that works.

§11. OFDM

⚠️ Why nearly everything modern — Wi-Fi, LTE, 5G, DVB — uses it. ⚠️ The problem it solves: at high data rates, symbols become shorter than the multipath delay spread, so echoes of one symbol smear into the next — inter-symbol interference — and equalizing that in a single wideband carrier is brutally hard. ⚠️ The solution: split the channel into many narrow, orthogonal subcarriers, each carrying a low-rate stream. ⚠️ Each subcarrier's symbol is now long compared to the delay spread, and a CYCLIC PREFIX (guard interval) absorbs the remaining echo. ⚠️ And it's cheap because it's an FFT. Modulation and demodulation are IFFT/FFT operations, which is precisely why OFDM became practical when DSP got fast enough. ⚠️ The costs: high peak-to-average power ratio (PAPR), demanding linear amplifiers and hurting efficiency — ⚠️ which is why uplinks sometimes use SC-FDMA instead — and sensitivity to frequency offset and phase noise. OFDMA (Wi-Fi 6+, LTE) — ⚠️ allocates subsets of subcarriers to different users, so several small devices share one transmission instead of each taking a full turn. This is a capacity/efficiency feature, not a speed feature.

§12. Coding and Retransmission

⚠️ Every real radio link is running errors and correcting them beneath you.

FEC    ⚠️ forward error correction — add redundancy so errors are fixed
       without retransmission. Convolutional/Viterbi, Reed-Solomon,
       ⚠️ Turbo and LDPC (near-Shannon; LDPC in Wi-Fi, 5G), Polar (5G control)
CRC    ⚠️ DETECTS errors; does not correct them
INTERLEAVING  ⚠️ scatter bits in time so a BURST error becomes many
       single-bit errors that FEC can handle. Essential against fading (§6)
ARQ    retransmission. ⚠️ HARQ combines retransmissions with the failed
       copy rather than discarding it — used throughout LTE/5G

⚠️ Coding gain is real and large: a good FEC can buy 5–10 dB, which by §3 → radio-intuitions-spectrum-link-budget-and-tradeoffs is several times the range. ⚠️ It costs data rate and latency, which is why low-latency modes use weaker coding.

§13. Multiple Access

FDMA / TDMA / CDMA / OFDMA / SDMA   ⚠️ divide by frequency, time, code,
   subcarrier, or space (beamforming/MIMO)
⚠️ CSMA/CA   listen before transmit, random backoff — Wi-Fi. ⚠️ Note
   CSMA/CD (collision DETECTION, from Ethernet) is IMPOSSIBLE on radio:
   you cannot hear a collision while transmitting
ALOHA        ⚠️ just transmit and hope. LoRaWAN. Simple; capacity collapses
   under load — classic ALOHA tops out around 18% channel utilization

⚠️ GOTCHA — the hidden node problem is the classic wireless-only failure and it has no wired analogue. ⚠️ A and C can both hear B but not each other, so both sense the channel as clear and both transmit, colliding at B. Carrier sensing cannot fix this — RTS/CTS can, at a cost in overhead. Symptom: throughput collapses when a particular pair of nodes is active, and each node's local view looks fine.

PART II — SOFTWARE-DEFINED RADIO

§14. ⚠️ SDR and IQ Sampling

⚠️ The concept that makes radio tractable for software people, and the one most tutorials explain badly.

        ┌─────────┐   ┌──────┐   ┌─────┐   ┌──────────────┐
RF ────►│ LNA/Mix │──►│ Filt │──►│ ADC │──►│ YOUR SOFTWARE│
        └─────────┘   └──────┘   └─────┘   └──────────────┘
        ⚠️ hardware shrinks; software grows

⚠️ IQ (complex baseband) is the key idea. The radio mixes the signal down so the carrier is at 0 Hz, producing two streams: I (in-phase) and Q (quadrature, 90° shifted). ⚠️ Together they form a COMPLEX number per sample. ⚠️ Why complex, concretely:

• ⚠️ A real-valued signal cannot distinguish +10 kHz from −10 kHz relative to the carrier. The complex representation can, so you get the full spectrum around your tuned frequency rather than a folded-over version.
• ⚠️ Amplitude = |I + jQ|; phase = atan2(Q, I). Every modulation in §8 → radio-antennas-propagation-noise-and-modulation becomes arithmetic on complex numbers.
• A constellation diagram is literally a scatter plot of your IQ samples.
• ⚠️ Sample rate = the bandwidth you can see. A 2 Msps complex stream gives you 2 MHz of spectrum, centred on your tuning frequency.

⚠️ Practical gotchas that will confuse you first time:

• ⚠️ DC spike at centre frequency — an artefact of direct-conversion receivers (LO leakage), not a real signal. Tune slightly off-target to avoid burying your signal under it.
• ⚠️ IQ imbalance produces mirror images of real signals reflected about the centre.
• ⚠️ Automatic gain control will lie to you about absolute power.
• ⚠️ Overflow/dropped samples: if your processing can't keep up, samples are simply lost, and the symptom is corrupt demodulation rather than an error message.

§15. Sampling and DSP Essentials

⚠️ NYQUIST      sample rate must exceed 2× the signal BANDWIDTH.
   ⚠️ For complex/IQ sampling, sample rate = bandwidth (not 2×) —
   the complex representation already carries the sign of frequency
ALIASING        ⚠️ under-sampled content folds back and is indistinguishable
   from real signal. Anti-alias filtering is mandatory, not optional
DECIMATION      ⚠️ filter THEN downsample. Downsampling without filtering
   first aliases — the single most common beginner DSP bug
INTERPOLATION   upsample and filter
FFT             ⚠️ time → frequency. Bin width = sample rate / FFT size.
   ⚠️ WINDOWING (Hann, Blackman) reduces spectral leakage from the implicit
   rectangular window; without it a strong tone smears across bins
FILTERS         FIR (⚠️ linear phase, stable, more taps) vs IIR (efficient,
   phase distortion). ⚠️ For radio, linear phase usually matters

⚠️ The receive chain in software, in order: tune → sample (IQ) → filter to the signal's bandwidth → decimate → correct frequency offset → time-synchronize → equalize → demodulate to symbols → decode (FEC) → deframe → CRC. ⚠️ Synchronization is where most of the difficulty lives. Your oscillator and theirs disagree — carrier frequency offset, sample timing offset, and phase — and estimating and tracking those is the bulk of a real demodulator.

§16. SDR Toolchain

⚠️ RTL-SDR       ~$30, RX only, ~24–1766 MHz, 8-bit, ~2.4 MHz BW.
   ⚠️ Start here. A repurposed TV tuner and genuinely capable for learning
HackRF One       ~$300, TX+RX (half duplex), 1 MHz–6 GHz, 8-bit, 20 MHz
LimeSDR / PlutoSDR  mid-range, full duplex, better ADCs
USRP (Ettus)     ⚠️ research/production grade, expensive, excellent
BladeRF          mid-high

SOFTWARE
⚠️ GNU Radio     flowgraph-based DSP framework. The standard. Python/C++ blocks
SoapySDR         ⚠️ hardware abstraction — write once, swap radios
GQRX / SDR++ / SDRangel   general-purpose receivers and spectrum viewing
Inspectrum       ⚠️ excellent for visually reverse-engineering a capture
Universal Radio Hacker  ⚠️ purpose-built for protocol reverse engineering
liquid-dsp / NumPy+SciPy   ⚠️ roll your own; NumPy is fine for offline work
srsRAN / OpenAirInterface  open-source LTE/5G stacks

⚠️ A learning path that actually works: receive FM broadcast (proves the chain) → decode ADS-B aircraft transponders at 1090 MHz (real digital demodulation, instant visible results) → decode a 433 MHz sensor around your home (OOK, simple, and it teaches framing) → then attempt transmit — carefully, and see §23 → radio-regulatory-security-and-debugging first.

PART III — PROTOCOLS