Why Diversity Works

The Fading Problem

Over an AWGN channel, the BER for BPSK decays exponentially with SNR: Pb=Q(2Ξ³)∼eβˆ’Ξ³P_b = Q(\sqrt{2\gamma}) \sim e^{-\gamma}. Over a Rayleigh fading channel, however, deep fades β€” events where ∣h∣2β‰ˆ0|h|^2 \approx 0 β€” occur with non-negligible probability. The resulting BER decays only as Pbβ‰ˆ1/(4Ξ³Λ‰)P_b \approx 1/(4\bar{\gamma}): inversely with SNR rather than exponentially. No amount of transmit power can overcome this fundamental limitation with a single fading link.

The solution is diversity: obtain multiple independently faded copies of the signal, so the probability that all copies simultaneously suffer a deep fade becomes vanishingly small.

Definition:

Diversity

Diversity is the technique of providing the receiver with multiple independently faded replicas of the same information-bearing signal. These replicas may be obtained across:

  • Space (multiple antennas): spatial diversity
  • Time (repeated transmission with sufficient delay): temporal diversity
  • Frequency (transmission over separated frequency bands): frequency diversity

The key requirement is independence: the fading on each replica must be approximately independent so that a deep fade on one branch is unlikely to coincide with deep fades on the others.

Independence is typically achieved by separating the replicas by more than the coherence distance (space), coherence time (time), or coherence bandwidth (frequency) of the channel.

Definition:

Diversity Order

The diversity order (or diversity gain) of a communication system is defined as

d=βˆ’lim⁑SNRβ†’βˆžlog⁑Pe(SNR)log⁑SNRd = -\lim_{\text{SNR} \to \infty} \frac{\log P_e(\text{SNR})}{\log \text{SNR}}

where PeP_e is the error probability (BER or SER).

A system with diversity order dd has error probability that decays as Pe∝SNRβˆ’dP_e \propto \text{SNR}^{-d} at high SNR. Equivalently, on a log-log plot of PeP_e vs SNR, the curve has slope βˆ’d-d in the high-SNR regime.

For a single Rayleigh fading link, d=1d = 1. With LL independent branches and optimal combining, d=Ld = L.

Diversity order captures the slope of the error probability curve at high SNR, not its vertical position. Two systems can have the same diversity order but different coding gains (horizontal shifts).

,

Theorem: BER Scaling with Diversity Order

Consider a system with LL independent Rayleigh fading branches, each with average SNR Ξ³Λ‰\bar{\gamma}. With optimal combining (maximal-ratio combining), the average BER for BPSK scales as

Pbβ‰ˆ(2Lβˆ’1L)1(4Ξ³Λ‰)LasΒ Ξ³Λ‰β†’βˆžP_b \approx \binom{2L-1}{L} \frac{1}{(4\bar{\gamma})^L} \quad \text{as } \bar{\gamma} \to \infty

The diversity order is d=Ld = L: each additional independent branch adds one to the exponent, steepening the BER vs SNR slope by one decade per 10 dB of additional SNR.

A deep fade occurs when the channel gain ∣h∣2|h|^2 falls below a threshold of order 1/Ξ³Λ‰1/\bar{\gamma}. For Rayleigh fading, P(∣h∣2<x)β‰ˆxP(|h|^2 < x) \approx x for small xx. With LL independent branches, the probability that all branches simultaneously fade is Pβ‰ˆ(1/Ξ³Λ‰)L=Ξ³Λ‰βˆ’LP \approx (1/\bar{\gamma})^L = \bar{\gamma}^{-L}.

Proof
Single-branch BER

For a single Rayleigh branch with SNR γ∼Exp(Ξ³Λ‰)\gamma \sim \text{Exp}(\bar{\gamma}), the average BER for BPSK is

Pb=12(1βˆ’Ξ³Λ‰1+Ξ³Λ‰)β‰ˆ14Ξ³Λ‰(γˉ≫1)P_b = \frac{1}{2}\left(1 - \sqrt{\frac{\bar{\gamma}}{1+\bar{\gamma}}}\right) \approx \frac{1}{4\bar{\gamma}} \quad (\bar{\gamma} \gg 1)

This gives diversity order d=1d = 1.

L-branch MRC SNR distribution

With MRC, the output SNR is Ξ³MRC=βˆ‘l=1LΞ³l\gamma_{\text{MRC}} = \sum_{l=1}^{L} \gamma_l. For i.i.d. Rayleigh branches, each Ξ³l∼Exp(Ξ³Λ‰)\gamma_l \sim \text{Exp}(\bar{\gamma}), so Ξ³MRC\gamma_{\text{MRC}} has the Erlang(L,Ξ³Λ‰)(L, \bar{\gamma}) distribution:

fΞ³MRC(Ξ³)=Ξ³Lβˆ’1eβˆ’Ξ³/Ξ³Λ‰Ξ³Λ‰L(Lβˆ’1)!,Ξ³β‰₯0f_{\gamma_{\text{MRC}}}(\gamma) = \frac{\gamma^{L-1} e^{-\gamma/\bar{\gamma}}} {\bar{\gamma}^L (L-1)!}, \quad \gamma \geq 0

Average BER with MRC

Averaging the conditional BER Q(2Ξ³)Q(\sqrt{2\gamma}) over the Erlang distribution and applying the high-SNR approximation yields

Pbβ‰ˆ(2Lβˆ’1L)1(4Ξ³Λ‰)LP_b \approx \binom{2L-1}{L} \frac{1}{(4\bar{\gamma})^L}

The exponent LL confirms diversity order d=Ld = L. β– \blacksquare

,

Example: Diversity Gain at 20 dB SNR

Compare the BER of BPSK at Ξ³Λ‰=20\bar{\gamma} = 20 dB (= 100 linear) for: (a) no diversity (L=1L = 1), (b) L=2L = 2, (c) L=4L = 4 branches with MRC over i.i.d. Rayleigh fading.

Solution
No diversity ($L = 1$)

Pbβ‰ˆ14Ξ³Λ‰=1400=2.5Γ—10βˆ’3P_b \approx \frac{1}{4\bar{\gamma}} = \frac{1}{400} = 2.5 \times 10^{-3}$

2-branch MRC ($L = 2$)

Pbβ‰ˆ(32)1(4Γ—100)2=3160000=1.875Γ—10βˆ’5P_b \approx \binom{3}{2} \frac{1}{(4 \times 100)^2} = \frac{3}{160000} = 1.875 \times 10^{-5}AnimprovementofmorethantwoordersofmagnitudeoverAn improvement of more than two orders of magnitude overL = 1$.

4-branch MRC ($L = 4$)

Pbβ‰ˆ(74)1(400)4=352.56Γ—1010=1.37Γ—10βˆ’9P_b \approx \binom{7}{4} \frac{1}{(400)^4} = \frac{35}{2.56 \times 10^{10}} = 1.37 \times 10^{-9}MorethansixordersofmagnitudebetterthanMore than six orders of magnitude better thanL = 1.Theimprovementisdramatic:eachdoublingof. The improvement is dramatic: each doubling ofLroughlysquarestheimprovementfactorathighSNR.roughly squares the improvement factor at high SNR.\blacksquare$

Diversity Order Visualization

Compare BER vs SNR curves for different numbers of diversity branches. On the log-log scale, the slope of each curve at high SNR equals the diversity order LL. Observe how additional branches steepen the curve.

Parameters

Fading With and Without Diversity

Visualise the received signal amplitude and instantaneous SNR across a sequence of transmitted symbols. Without diversity, deep fades cause bursts of errors. With diversity combining, the combined SNR is stabilised and deep fades become rare.

Parameters
1
15
500

Quick Check

A communication system achieves BER =10βˆ’3= 10^{-3} at SNR =20= 20 dB and BER =10βˆ’6= 10^{-6} at SNR =30= 30 dB. What is the approximate diversity order?

d=1d = 1

d=2d = 2

d=3d = 3

d=6d = 6

Correction:
d=3d = 3

A 10 dB (factor of 10) increase in SNR produces a 10310^3 (three decades) improvement in BER: Pe∝SNRβˆ’3P_e \propto \text{SNR}^{-3}, so d=3d = 3.

Common Mistake: Diversity Requires Independence

Mistake:

Assuming that adding more antennas always provides more diversity, regardless of their spacing or the propagation environment.

Correction:

Diversity gain comes from independent fading across branches. If antennas are spaced too closely (less than about Ξ»/2\lambda/2 in a rich-scattering environment), the fading becomes correlated and the effective diversity order is reduced. In a line-of-sight channel with no scattering, spatial diversity provides almost no benefit regardless of antenna spacing, because all paths experience the same fade.

The rule of thumb for spatial diversity: antenna spacing β‰₯Ξ»/2\geq \lambda/2 at the mobile and β‰₯10Ξ»\geq 10\lambda at the base station (due to narrower angular spread at elevated antennas).

Historical Note: Diversity in Early Radio

1927-1959

The concept of diversity reception dates back to the 1920s and 1930s, when shortwave radio operators discovered that using multiple antennas separated by several wavelengths dramatically improved reception reliability. H. H. Beverage and H. O. Peterson at RCA (1931) published one of the first systematic studies of space diversity for transoceanic radio links, showing that fading on separated antennas was largely uncorrelated. By the 1950s, diversity combining was standard practice in tropospheric scatter and microwave relay systems. The mathematical foundations were laid by Brennan (1959), who analysed selection, equal-gain, and maximal-ratio combining β€” the same three techniques still central to modern wireless systems.

Diversity Order β€” BER Curves Steepening

Watch the BER vs SNR curves steepen as diversity branches are added. Each additional branch increases the high-SNR slope by one decade per 10 dB, transforming the error probability from inverse-linear to inverse-polynomial in SNR.
BER curves for L=1,2,4,8L = 1, 2, 4, 8 branches with MRC over i.i.d. Rayleigh fading. The slope at high SNR equals the diversity order LL.

Key Takeaway

Diversity changes the slope of the BER vs SNR curve. Without diversity, BER in Rayleigh fading decays as 1/SNR1/\text{SNR} (diversity order 1). With LL independent branches and optimal combining, BER decays as 1/SNRL1/\text{SNR}^{L} (diversity order LL). This transforms the error probability from an inverse-linear to an inverse-polynomial function of SNR, providing enormous reliability gains at high SNR.

Diversity Order

The negative slope of the log-log BER vs SNR curve at high SNR. A system with diversity order dd achieves Pe∝SNRβˆ’dP_e \propto \text{SNR}^{-d}. Also called the diversity degree or diversity exponent.

Related: Diversity Gain, Maximal-Ratio Combining (MRC), Rayleigh Distribution

Diversity Gain

The reduction in required SNR to achieve a target BER, obtained by exploiting multiple independently faded signal copies. Often quantified relative to a single-branch system at the same BER.

Related: Diversity Order, Selection Combining (SC), Array Gain

Independent Fading

A condition where the fading coefficients on different diversity branches are statistically independent. This requires sufficient separation in space (more than coherence distance), time (more than coherence time), or frequency (more than coherence bandwidth).

Related: Coherence Time, Coherence Time, Spatial Diversity and MIMO Detection

PrerequisitesReceive Diversity