Ferkans — Interactive Telecom Tutor

Why Frequency Diversity Needs Coding

The point is that a frequency-selective channel with $L$ resolvable paths offers an intrinsic frequency diversity resource. Plain OFDM converts the selective channel into $N_{\rm sc}$ parallel flat sub-channels — and in doing so throws the diversity away: each subcarrier experiences only one complex gain. A coded modulation system that spreads coded bits across subcarriers with a good interleaver recovers that diversity — up to the minimum Hamming distance of the outer code. This is the BICM-OFDM design pattern that underpins every modern wireless standard.

Definition:
OFDM Transmission

An OFDM transmitter carrying frequency-domain symbols $\{X[k]\}_{k=0}^{N_{\rm sc}-1}$ produces time-domain samples $x[n] = \frac{1}{\sqrt{N_{\rm sc}}} \sum_{k=0}^{N_{\rm sc}-1} X[k]\, e^{j 2\pi nk / N_{\rm sc}}, \quad n = 0, \ldots, N_{\rm sc}-1,$ followed by the cyclic prefix of $N_{\rm cp}$ samples prepended from the tail of the block. After the channel, the receiver removes the prefix and applies the inverse DFT to obtain $Y[k] = H[k]\,X[k] + W[k], \quad k = 0, \ldots, N_{\rm sc}-1,$ where $H[k]$ is the channel's frequency response at subcarrier $k$ .

Theorem: OFDM as Parallel Flat-Fading Channels

Let the discrete-time channel be $h[n] = \sum_{\ell=0}^{L-1} h_\ell\,\delta[n-\ell]$ with $L-1 \le N_{\rm cp}$ . Then, after CP removal and the $N_{\rm sc}$ -point DFT, the OFDM output equals $Y[k] = H[k] X[k] + W[k], \qquad H[k] = \sum_{\ell=0}^{L-1} h_\ell\, e^{-j 2\pi k\ell / N_{\rm sc}}$ with $W[k] \sim \mathcal{CN}(0, \sigma^2)$ i.i.d. across $k$ .

Proof

Cyclic-prefix convolution

With $N_{\rm cp} \ge L-1$ , the linear channel convolution between the CP-appended block and $h[n]$ becomes circular over the useful OFDM block of length $N_{\rm sc}$ .

DFT diagonalises circular convolution

The DFT matrix $\mathbf{F}$ diagonalises circular convolution: $\mathbf{F}\,(\mathbf{h} \circledast \mathbf{x}) = \mathrm{diag}(\mathbf{F}\mathbf{h}) \cdot \mathbf{F}\mathbf{x}$ . The eigenvalues are $H[k] = \sum_\ell h_\ell e^{-j 2\pi k\ell / N_{\rm sc}}$ .

Noise remains white after DFT

Since $\mathbf{F}$ is unitary, $\mathbf{F}\mathbf{w} \sim \mathcal{CN}(0, \sigma^2 \mathbf{I})$ , so $W[k]$ is still i.i.d. complex Gaussian.

Result

Putting the three steps together: $Y[k] = H[k] X[k] + W[k]$ — a parallel set of $N_{\rm sc}$ scalar flat-fading channels, one per subcarrier. $\blacksquare$

,

Why Uncoded OFDM Throws Away the Diversity

Each subcarrier now sees a SINGLE fading coefficient $H[k]$ . Uncoded detection per subcarrier has diversity order 1 — no different from a flat-fading scalar channel. The $L$ paths of the original channel are HIDDEN in the frequency-domain structure of $\{H[k]\}$ , but not exploited. To recover them, one must SPREAD the code's parity across subcarriers so that each bit error involves multiple subcarriers.

Definition:
Frequency Diversity

The frequency diversity $d_f$ of a frequency-selective channel with $L$ resolvable paths is the number of statistically independent fades available across the system bandwidth. For an $L$ -tap channel with uncorrelated taps, $d_f = L$ . A code that spreads every bit-error event across $\ge L$ subcarriers can achieve diversity order $L$ ; codes with smaller spread are limited by the code's Hamming distance.

BICM-OFDM BER: Frequency Diversity $\min(d_H, L)$

Adjust the channel path count $L$ and the code's minimum Hamming distance $d_H$ . The diversity order of BICM-OFDM is the minimum of the two: when the code is stronger than the channel, $L$ dominates; when the channel has more paths than the code has distance, $d_H$ is the bottleneck.

Parameters

Channel paths

L

4

Code

d_H

5

Example: Frequency Diversity in ITU-R Vehicular-B

The ITU-R Vehicular-B channel model has $L = 7$ resolvable paths spanning $\tau_{\max} = 20\,\mu\mathrm{s}$ . With a 5 MHz OFDM signal and a rate-1/2 convolutional code of constraint length $K = 7$ ( $d_H = 10$ ), what diversity order does BICM-OFDM achieve? What if we instead use a very weak code with $d_H = 4$ ?

Solution

Strong code ($d_H = 10$)

$d = \min(d_H, L) = \min(10, 7) = 7$ : we harvest the full frequency diversity of the channel. The code is stronger than the channel can reward; the bottleneck is physical.

Weak code ($d_H = 4$)

$d = \min(4, 7) = 4$ : the code leaks diversity. The channel still offers 7 independent fades, but only 4 bits separate any pair of codewords, so only 4 independent fade events appear in the union bound.

Design implication

For this channel, any code with $d_H \ge 7$ extracts the full frequency diversity. Going beyond $d_H = 7$ improves coding gain (the multiplicative constant) but not the diversity slope.

Common Mistake: Narrowband Channels Have No Frequency Diversity to Harvest

Mistake:

Assuming BICM-OFDM always gives diversity $d_H$ : "our code has $d_H = 20$ , so we get diversity 20."

Correction:

Only $\min(d_H, L)$ is available. On a narrowband channel with $L = 1$ , OFDM trivialises to a flat-fading scalar channel, and BICM-OFDM has diversity 1 no matter how large $d_H$ is. Frequency diversity requires frequency selectivity — the bandwidth must exceed the coherence bandwidth.

⚠️Engineering Note

OFDM Is Ubiquitous in Wireless Standards

BICM-OFDM underpins every modern wireless standard for broadband communication:

4G LTE: LDPC is being phased in (NR), TBCC legacy; BICM-OFDM on 15 kHz subcarriers.
5G NR: LDPC + BICM-OFDM with subcarrier spacings 15/30/60/120/240 kHz (flexible numerology).
Wi-Fi 6 (802.11ax): LDPC + BICM-OFDM + OFDMA.
DVB-T2: LDPC + BICM-OFDM with rotated constellations.
ATSC 3.0: LDPC + BICM-OFDM + non-uniform constellations.

Practical Constraints

•
Rate matching via puncturing + CCDM enables continuous rate
•
Pilot overhead 5-10% for channel estimation

,

Historical Note: Origins of OFDM: Chang 1966, Weinstein-Ebert 1971

OFDM was proposed by R. W. Chang (1966, AT&T Bell Labs) for efficient frequency-division multiplexing. The digital FFT-based implementation was shown feasible by Weinstein and Ebert in 1971, but OFDM remained largely impractical until DSP hardware caught up in the late 1980s. Its first mass deployment was in DAB (Digital Audio Broadcasting, 1995), then ADSL (1995) and 802.11a Wi-Fi (1999). Today OFDM is the dominant waveform for broadband wireless — a paradigm shift from the single-carrier equalisers it displaced.

Key Takeaway

OFDM converts a frequency-selective channel with $L$ paths into $N_{\rm sc}$ parallel scalar flat channels. Per-subcarrier detection loses the diversity; BICM across subcarriers recovers it, up to the bottleneck $\min(d_H, L)$ .

BICM-OFDM: Harvesting Frequency Diversity