Why does BICM-ID require non-Gray mapping?

Gray mapping assigns bit labels to constellation points such that adjacent points differ by exactly one bit. This minimizes BER for standard (non-iterative) BICM because most symbol errors cause only one bit error. However, Gray mapping creates a fundamental problem for iterative decoding: the demapper cannot generate useful extrinsic information. In information-theoretic terms, the mutual information between the demapper output and the coded bits, conditioned on the a priori information from the decoder, equals zero for Gray-mapped constellations at low a priori quality. The demapper's EXIT characteristic is flat, meaning no matter how much feedback the decoder provides, the demapper cannot improve its output. Non-Gray mappings (anti-Gray, set-partitioning, or random mapping) create bit labels where adjacent constellation points can differ by multiple bits. This means the a priori information about some bits significantly changes the demapper's LLR computation for other bits within the same symbol, generating non-zero extrinsic information. For example, with 16-QAM and set-partitioning mapping: knowing bit b0 with high confidence changes the effective constellation from 16 points to 8 points for the remaining bits, substantially improving the LLR quality. The EXIT characteristic rises steeply, enabling iterative convergence.

How does the iterative decoding loop work?

The BICM-ID receiver operates as a turbo-like iterative loop between the soft demapper and the soft decoder. Iteration 0 (initial): The soft demapper computes LLRs for each bit position using only the received symbol y and the channel state h. No a priori information is available: L_a(b_i) = 0 for all coded bits. The LLRs are deinterleaved and passed to the decoder. Iteration k (k >= 1): (1) The decoder processes the LLRs from the demapper and produces soft outputs (posterior LLRs) for each coded bit. The extrinsic information is extracted: L_e(b_i) = L_posterior(b_i) - L_input(b_i). This extrinsic information represents what the code structure has learned about each bit beyond what the channel observation provided. (2) The extrinsic LLRs are interleaved and fed back to the demapper as a priori information: L_a(b_i) = interleaved L_e(b_i). (3) The demapper recomputes the bit LLRs using both the received symbol and the a priori information: L_demapper(b_i) = log[sum_{s: b_i=1} P(y|s,h) * prod_{j!=i} P(b_j=label_j(s)))] - log[sum_{s: b_i=0} P(y|s,h) * prod_{j!=i} P(b_j=label_j(s)))]. The a priori probabilities P(b_j) from the decoder feedback effectively reduce the constellation size for computing each bit's LLR.

What is EXIT chart analysis for BICM-ID?

EXIT (EXtrinsic Information Transfer) chart analysis is a semi-analytical tool for predicting the convergence behavior of iterative receivers without running full Monte Carlo simulations. For BICM-ID, two EXIT characteristics are plotted: (1) Demapper EXIT curve: plots the mutual information at the demapper output (I_E,demapper) versus the mutual information of the a priori input (I_A,demapper) at a given SNR. This curve depends on the constellation mapping (Gray, anti-Gray, set-partitioning), the modulation order, and the SNR. For Gray mapping: the curve is nearly flat (I_E approximately constant regardless of I_A), confirming zero iterative gain. For anti-Gray mapping: the curve rises steeply from I_A = 0 to I_A = 1, indicating strong iterative gain. (2) Decoder EXIT curve: plots the mutual information at the decoder output (I_E,decoder) versus the mutual information of the input (I_A,decoder). This curve depends only on the code (rate, constraint length, type) and is independent of the channel. For convergence: the demapper EXIT curve must lie above the (mirrored) decoder EXIT curve for all values of mutual information from 0 to 1. If the curves intersect before reaching I_E = 1, the iterative process will stall at that mutual information level, resulting in a BER floor. The 'tunnel' between the two curves must remain open for reliable convergence. A wider tunnel indicates faster convergence and more robust performance. Design procedure: (1) Choose a target SNR. (2) Plot the demapper EXIT curve for different mappings at that SNR. (3) Overlay the decoder EXIT curve.

Digital Comms / Iterative

BICM-ID

/BIK-um eye-DEE/ — with Iterative Decoding

Extension of BICM with a decoder-to-demapper feedback loop that iteratively refines bit LLRs. Requires non-Gray constellation mapping to generate extrinsic information. Recovers 0.5–1.5 dB of AWGN capacity gap vs. standard BICM. Within 0.3 dB of coded modulation capacity after 8–10 iterations. EXIT chart analysis designs mapping and code rate for convergence. Standard in DVB-S2, DVB-T2.

Iterations: 8–10 typical

Gap to capacity: <0.3 dB

Mapping: Non-Gray required

Understanding BICM-ID

Standard BICM with Gray mapping leaves 0.5–1.5 dB on the table relative to the coded modulation capacity on AWGN channels. BICM-ID recovers this gap by treating the demapper and decoder as two constituent components of a turbo-like iterative system. The decoder provides a priori information about coded bits back to the demapper, which uses it to refine its LLR estimates, particularly for the less-protected bit positions in high-order constellations (64-QAM, 256-QAM).

The critical design constraint is that Gray mapping generates zero extrinsic information at the demapper (its EXIT curve is flat), making iteration pointless. Non-Gray mappings (anti-Gray, set-partitioning, or random) create bit-label structures where knowledge of some bits significantly improves the demapper's LLR for other bits within the same symbol, enabling iterative convergence.

Iterative Loop Structure

Demapper with A Priori:
L(b_i) = log[Σ_{s:b_i=1} P(y|s,h)∏_j≠iP(b_j)] − log[Σ_{s:b_i=0} P(y|s,h)∏_j≠iP(b_j)]

Extrinsic Information:
L_e(b_i) = L_posterior(b_i) − L_input(b_i)

Convergence:
Demapper EXIT I_E vs. I_A must stay above decoder EXIT
Open tunnel → reliable convergence to I_E = 1
Closed tunnel → BER floor

Mapping Strategy Comparison

Mapping	1st Iteration BER	Iterative Gain	Converged BER	Use Case
Gray	Best	None (flat EXIT)	Same as 1st	Standard BICM
Anti-Gray	Worst	Maximum	Best after 8–10 iter	BICM-ID optimal
Set-partitioning	Moderate	Good	Near anti-Gray	DVB-S2
Mixed/random	Moderate	Moderate	Good	Flexible designs

Performance vs. Complexity

Iterations	AWGN Gap	Complexity	Latency
1 (standard BICM)	0.5–1.5 dB	1×	Lowest
4	0.5–0.8 dB	~8×	Moderate
8–10	<0.3 dB	~16–20×	High
20+	<0.1 dB	~40×	Very high

Common Questions

Frequently Asked Questions

Why non-Gray mapping?

Gray mapping: adjacent points differ by 1 bit, so a priori knowledge of other bits doesn't change the demapper's LLR (flat EXIT curve, zero iterative gain). Non-Gray: adjacent points differ by multiple bits, so decoder feedback about some bits reshapes the effective constellation for remaining bits. Anti-Gray: maximum iterative gain but worst first-iteration BER.

Iterative loop operation?

Decoder extrinsic LLRs → interleave → demapper a priori. Demapper recomputes LLRs using received symbol + a priori from decoder. New extrinsic → deinterleave → decoder. 8–10 iterations: cost ~16–20× single-pass, within 0.3 dB of capacity.

EXIT chart design?

Plot demapper EXIT (I_E vs. I_A) at target SNR alongside (mirrored) decoder EXIT. Open tunnel between curves = reliable convergence. Closed = BER floor. Choose mapping and code rate for just-open tunnel at target SNR. DVB-S2 LDPC + mapping jointly optimized via EXIT analysis.

Related Terms

← BICM Detail BiCMOS Process →

Digital Communications

Precision RF Components

RF Essentials provides precision terminations and custom RF assemblies for iterative receiver testing, BER measurement systems, and coded modulation performance validation.

Request a Quote