How is the Bhattacharyya parameter defined?

For a binary-input discrete memoryless channel W with input alphabet {0,1} and output alphabet Y, the Bhattacharyya parameter is Z(W) = sum over all y in Y of sqrt(W(y|0) * W(y|1)). This is the sum of the geometric mean of the two conditional output distributions. Intuitively, Z measures how much the output distributions overlap: if W(y|0) and W(y|1) are identical for all y, then Z = 1 (outputs carry no information about the input). If the output distributions have disjoint support (no y is produced by both inputs), then Z = 0 (outputs perfectly identify the input). The Bhattacharyya parameter upper-bounds the maximum-likelihood decoding error probability for a single binary decision: P(error) <= Z(W)/2. For coded systems with minimum Hamming distance d_min, the pairwise error probability between two codewords at distance d is bounded by P_pair <= Z^d. The union bound over all codeword pairs gives the block error probability bound: P_block <= sum over all non-zero codewords of Z^(weight of codeword), which for a linear code with weight enumerator A(d) becomes P_block <= sum_{d=d_min}^{n} A(d) * Z^d. This is the Bhattacharyya bound, which is computationally simple and reasonably tight at moderate to high SNR.

How is the Bhattacharyya parameter used in polar codes?

Polar codes, invented by Erdal Arikan in 2009, use channel polarization to transform N copies of a channel W into N synthetic sub-channels W_N^(i), where i = 1 to N. As N grows (N = 2^n), the Bhattacharyya parameters of these sub-channels polarize: a fraction approaching the channel capacity I(W) will have Z(W_N^(i)) approaching 0 (perfectly reliable), and the remaining fraction (1 - I(W)) will have Z(W_N^(i)) approaching 1 (completely unreliable). To construct a polar code of rate R: compute Z(W_N^(i)) for all N sub-channels using the recursive relations Z(W_N^(2i-1)) = 2*Z(W_N/2^(i)) - Z(W_N/2^(i))^2 (upper channel, less reliable) and Z(W_N^(2i)) = Z(W_N/2^(i))^2 (lower channel, more reliable). Sort sub-channels by Z value. Select the K = N*R sub-channels with the smallest Z values as data channels. Set the remaining N-K sub-channels as frozen channels (transmit known values, typically 0). The reliability ordering determined by Z values is the key design parameter of a polar code. The construction is channel-specific: for AWGN at a given Eb/N0, compute Z = exp(-Eb/N0) and propagate through the recursion. The resulting polar code is capacity-achieving as N approaches infinity, meaning the block error probability approaches 0 for any rate R < I(W).

Bhattacharyya parameter for common channels?

The Bhattacharyya parameter has closed-form expressions for several important channel models. Binary Symmetric Channel (BSC) with crossover probability p: Z(BSC) = 2*sqrt(p*(1-p)). At p = 0 (noiseless): Z = 0. At p = 0.5 (useless): Z = 1. At p = 0.1: Z = 0.6. Binary Erasure Channel (BEC) with erasure probability epsilon: Z(BEC) = epsilon. This is the simplest case because the BEC either delivers the correct symbol or declares an erasure. At epsilon = 0: Z = 0. At epsilon = 1: Z = 1. AWGN channel with BPSK modulation and energy per bit Eb: Z(AWGN) = exp(-Eb/N0) = exp(-SNR_b). At Eb/N0 = 0 dB (SNR = 1): Z = exp(-1) = 0.368. At Eb/N0 = 3 dB: Z = exp(-2) = 0.135. At Eb/N0 = 10 dB: Z = exp(-10) = 4.54 * 10^-5. The exponential decay of Z with SNR for AWGN explains why coded systems achieve very low error rates at moderate SNR: the pairwise error bound Z^d decreases doubly exponentially with distance d. For a code with d_min = 10 at Eb/N0 = 3 dB: P_pair <= (0.135)^10 = 2.0 * 10^-9. Rayleigh fading channel (no CSI at transmitter): Z(Rayleigh) = 1/(1 + Eb/N0). This decays only linearly with SNR (compared to exponentially for AWGN), reflecting the much higher error probability floor of uncoded fading channels and motivating the use of diversity techniques to convert the fading Z behavior toward the AWGN form.

Information Theory / Coding

Bhattacharyya Parameter

Q: How is the Bhattacharyya parameter used in polar codes?

Polar codes, invented by Erdal Arikan in 2009, use channel polarization to transform N copies of a channel W into N synthetic sub-channels W_N^(i), where i = 1 to N. As N grows (N = 2^n), the Bhattacharyya parameters of these sub-channels polarize: a fraction approaching the channel capacity I(W) will have Z(W_N^(i)) approaching 0 (perfectly reliable), and the remaining fraction (1 - I(W)) will have Z(W_N^(i)) approaching 1 (completely unreliable). To construct a polar code of rate R: compute Z(W_N^(i)) for all N sub-channels using the recursive relations Z(W_N^(2i-1)) = 2*Z(W_N/2^(i)) - Z(W_N/2^(i))^2 (upper channel, less reliable) and Z(W_N^(2i)) = Z(W_N/2^(i))^2 (lower channel, more reliable). Sort sub-channels by Z value. Select the K = N*R sub-channels with the smallest Z values as data channels. Set the remaining N-K sub-channels as frozen channels (transmit known values, typically 0). The reliability ordering determined by Z values is the key design parameter of a polar code. The construction is channel-specific: for AWGN at a given Eb/N0, compute Z = exp(-Eb/N0) and propagate through the recursion. The resulting polar code is capacity-achieving as N approaches infinity, meaning the block error probability approaches 0 for any rate R < I(W).

Q: Bhattacharyya parameter for common channels?

The Bhattacharyya parameter has closed-form expressions for several important channel models. Binary Symmetric Channel (BSC) with crossover probability p: Z(BSC) = 2*sqrt(p*(1-p)). At p = 0 (noiseless): Z = 0. At p = 0.5 (useless): Z = 1. At p = 0.1: Z = 0.6. Binary Erasure Channel (BEC) with erasure probability epsilon: Z(BEC) = epsilon. This is the simplest case because the BEC either delivers the correct symbol or declares an erasure. At epsilon = 0: Z = 0. At epsilon = 1: Z = 1. AWGN channel with BPSK modulation and energy per bit Eb: Z(AWGN) = exp(-Eb/N0) = exp(-SNR_b). At Eb/N0 = 0 dB (SNR = 1): Z = exp(-1) = 0.368. At Eb/N0 = 3 dB: Z = exp(-2) = 0.135. At Eb/N0 = 10 dB: Z = exp(-10) = 4.54 * 10^-5. The exponential decay of Z with SNR for AWGN explains why coded systems achieve very low error rates at moderate SNR: the pairwise error bound Z^d decreases doubly exponentially with distance d. For a code with d_min = 10 at Eb/N0 = 3 dB: P_pair <= (0.135)^10 = 2.0 * 10^-9. Rayleigh fading channel (no CSI at transmitter): Z(Rayleigh) = 1/(1 + Eb/N0). This decays only linearly with SNR (compared to exponentially for AWGN), reflecting the much higher error probability floor of uncoded fading channels and motivating the use of diversity techniques to convert the fading Z behavior toward the AWGN form.

/bah-tah-CHAR-yah/ — Z(W)

Statistical overlap metric between channel output distributions conditioned on binary inputs. Z(W) = ∑_y √(W(y|0) × W(y|1)). Bounds pairwise error: P_pair ≤ Z^d. For AWGN/BPSK: Z = e^−E_b/N₀. For BEC: Z = ε. Central to polar code construction: sub-channels with Z → 0 carry data; Z → 1 are frozen.

Range: 0 (noiseless) to 1 (useless)

AWGN: Z = e^−SNR_b

BEC: Z = ε

Understanding the Bhattacharyya Parameter

The Bhattacharyya parameter quantifies how well a channel decoder can distinguish between the two binary input symbols based on the channel output. A value near 0 means the output distributions are nearly disjoint (the channel is reliable), while a value near 1 means they overlap heavily (the channel is unreliable). This metric provides computationally simple upper bounds on decoding error probability and is the primary design tool for polar code construction.

For coded systems, the pairwise error probability between two codewords at Hamming distance d is bounded by Z^d, and the union bound over all codeword pairs gives the block error rate. The exponential decay of Z with SNR for AWGN channels explains why coded systems achieve very low error rates at moderate SNR: doubling the code distance squares the pairwise error bound.

Key Formulas

Definition:
Z(W) = ∑_y √(W(y|0) × W(y|1))

Pairwise Error Bound:
P_pair(d) ≤ Z^d
Block: P_block ≤ ∑_{d=d_min}ⁿ A(d) × Z^d

Channel-Specific Z:
AWGN/BPSK: Z = e^−E_b/N₀
BSC(p): Z = 2√(p(1−p))
BEC(ε): Z = ε
Rayleigh: Z = 1/(1 + E_b/N₀)

Polar Code Recursion:
Z⁽²ⁱ⁻¹⁾ = 2Z⁽ⁱ⁾ − (Z⁽ⁱ⁾)² (upper, less reliable)
Z⁽²ⁱ⁾ = (Z⁽ⁱ⁾)² (lower, more reliable)

Z Values for Common Channels

Channel	Parameter	Z	Decay with SNR
AWGN + BPSK	E_b/N₀	e^−E_b/N₀	Exponential
BSC	p = crossover	2√(p(1−p))	~Exponential
BEC	ε = erasure	ε	Linear
Rayleigh	E_b/N₀	1/(1+E_b/N₀)	Linear (1/SNR)

Numerical Examples

E_b/N₀	Z (AWGN)	Z¹⁰ (d_min=10)
0 dB	0.368	4.2 × 10⁻⁵
3 dB	0.135	2.0 × 10⁻⁹
6 dB	0.0183	4.4 × 10⁻¹⁸
10 dB	4.5 × 10⁻⁵	3.4 × 10⁻⁴⁴

Common Questions

Frequently Asked Questions

Definition?

Z(W) = ∑_y √(W(y|0)W(y|1)). Measures output distribution overlap. Z = 0: noiseless (disjoint outputs). Z = 1: useless (identical outputs). Bounds ML error: P_e ≤ Z/2. Pairwise: P_pair ≤ Z^d. Union bound: P_block ≤ ∑A(d)Z^d.

Polar code construction?

Channel polarization drives Z toward 0 or 1 for each sub-channel. Recursion: Z⁽²ⁱ⁻¹⁾ = 2Z − Z² (less reliable), Z⁽²ⁱ⁾ = Z² (more reliable). Select K sub-channels with smallest Z as data channels. Remainder frozen. Capacity-achieving as N → ∞. Channel-specific: AWGN Z = e^−E_b/N₀.

Z for common channels?

AWGN: e^−SNR (exponential decay). BSC: 2√(p(1−p)). BEC: ε (simplest). Rayleigh: 1/(1+SNR) (linear decay, motivates diversity). At E_b/N₀ = 3 dB, AWGN Z = 0.135; with d_min = 10: P_pair ≤ 2 × 10⁻⁹.

Related Terms

← BGA RF Bianisotropic Medium →

Coding Theory

Precision RF Components

RF Essentials provides precision terminations and custom RF assemblies for BER testing, channel coding validation, and modulation quality analysis in coded communication systems.

Request a Quote