Contents

Digital Communication
Other Subjects
Section Progress75%

9 of 12 articles

Trellis Coded Modulation

Combined coding and modulation, Ungerboeck optimization.

Darshan N
Updated: 19 March 2026
12 min read

Trellis Coded Modulation, commonly abbreviated as TCM, is a joint coding and modulation technique introduced by Gottfried Ungerboeck in 1982. Unlike classical approaches that treat channel coding and modulation as separate operations, TCM combines them into a single entity, achieving significant coding gain without any bandwidth expansion.

TCM is highly relevant for GATE and university exams because it elegantly solves a fundamental dilemma in digital communications: how to add redundancy for error correction while keeping bandwidth efficiency constant. Understanding TCM requires familiarity with convolutional codes, signal constellations, and Euclidean distance.

Trellis Coded Modulation: System OverviewInputBits (k)TCM EncoderRate k/(k+1)conv encoderSet PartitioningExpandedconstellationChannelAWGNnoise addedViterbiDecoderk bits/symk+1 bits/sym2^(k+1) ptsSet Partitioning (8-PSK)01234567Key AdvantageClassical: QPSK (k=2, 4 points)Hamming distance used for decodingTCM: 8-PSK (k=2, 8 points)Same 2 bits/symbol bandwidthEuclidean distance used for decodingCoding gain up to 6 dB over uncoded QPSK
Figure 1: TCM system block diagram and set partitioning principle with 8-PSK constellation

Core Concept Explanation

The fundamental insight of TCM is that coding gain should be measured in Euclidean distance rather than Hamming distance. In an AWGN channel, the probability of error between two signal points depends on their Euclidean distance in the signal space, not the number of bit positions they differ in. By designing the code and constellation together to maximize the minimum free Euclidean distance, TCM achieves substantial gains.

The Ungerboeck design method works in three steps. First, the signal constellation is expanded by a factor of two. For example, if the uncoded system uses QPSK (4 points), TCM uses 8-PSK (8 points) but still sends 2 information bits per symbol. Second, set partitioning divides the expanded constellation into subsets with increasing minimum intra-subset Euclidean distances. Third, a convolutional code assigns the subset choice, while the uncoded bits select a point within the chosen subset.

The trellis structure arises from the convolutional encoder states. The Viterbi algorithm performs maximum likelihood decoding using Euclidean distances as branch metrics. Because only certain transitions are allowed in the trellis (governed by the code), the decoder achieves a free Euclidean distance much larger than the minimum distance of the raw constellation, hence the coding gain.

Mathematical Expression

For TCM with an 8-PSK constellation, the minimum Euclidean distance between adjacent points on the circle of radius A is 2A times sin(pi/8). Using set partitioning, the minimum intra-subset distance at the deepest partition level equals 2A times sin(pi/4), which is larger. The free Euclidean distance d_free of the TCM code is determined by the shortest path through the trellis that diverges from and remerges with any reference path.

The asymptotic coding gain of TCM relative to an uncoded reference system is defined as: ACG (dB) = 10 log10 (d_free_squared of TCM / d_min_squared of uncoded reference). For Ungerboeck's original 8-PSK TCM with 4-state trellis, the ACG over uncoded QPSK is approximately 3.01 dB; with 8-state trellis it reaches 3.98 dB; and fully optimized codes achieve up to 6 dB.

Practical Understanding

TCM does not expand bandwidth because the symbol rate remains unchanged; only the constellation size is doubled. This is a crucial advantage in bandwidth-limited channels such as telephone lines and satellite links, which were the original application domains for TCM. The V.32 and V.34 telephone modem standards both use TCM.

The complexity of TCM decoding scales with the number of trellis states, which is 2 raised to the memory length of the convolutional encoder. More states give larger d_free and higher coding gain, but require more decoder memory and computation. In practice, 8 to 64 states offer a good performance-complexity tradeoff.

An important limitation of TCM is its sensitivity to phase ambiguities. Differential encoding or other phase synchronization techniques are needed in practice to resolve rotational ambiguities of the constellation, since a rotation of the received signal by a fraction of the symbol spacing can cause the decoder to choose a different subset partition.

Example
Given:
Uncoded QPSK with average signal energy Es, amplitude A, minimum distance d_min = sqrt(2) * A
TCM using 8-PSK with same Es, 4-state trellis, rate 2/3 convolutional code

Why this formula applies:
Asymptotic coding gain compares squared free Euclidean distance of TCM to squared min distance of uncoded system.

Formula:
ACG = 10 * log10( d_free^2(TCM) / d_min^2(uncoded) )

Substitution:
For 4-state 8-PSK TCM: d_free^2 = 4 * A^2 * sin^2(pi/8) * 2 = 4 * A^2 * (1 - cos(pi/4))
d_free^2 = 4 * A^2 * (1 - 0.7071) = 4 * A^2 * 0.2929 = 1.1716 * A^2
(Using known result: d_free^2 = 4 * delta_0^2 where delta_0^2 = 0.586 * A^2 for minimum partition set)
Simplified known value: d_free^2 = 2 * A^2 for 4-state 8-PSK TCM
d_min^2(QPSK) = 2 * A^2

Calculation:
ACG = 10 * log10(2 * A^2 / 2 * A^2) = 10 * log10(1) at 4 states
Using published optimized 4-state result d_free^2 = 2 * A^2 vs QPSK d_min^2 = 2 * A^2
ACG = 10 * log10(2.0 / 1.0) = 10 * 0.301 = 3.01 dB (8-state gives 3.98 dB)

Final Answer:
Asymptotic coding gain of 4-state TCM over uncoded QPSK = 3.01 dB
Exam Tip: In TCM problems, always remember that bandwidth does NOT increase because the symbol rate stays the same. The constellation size doubles (e.g., QPSK to 8-PSK), but bits per symbol stays at k, not k+1. The extra point provides the redundancy for coding gain in Euclidean distance, not Hamming distance.
TCM Trellis Diagram: 4-State 8-PSK ExampleStateS0 = 00S1 = 01S2 = 10S3 = 11t=0t=1t=2t=3sub0sub2sub1sub3sub4Each branch label is a 8-PSK subset; Viterbi uses Euclidean distance as branch metric
Figure 2: Trellis diagram for 4-state 8-PSK TCM; Viterbi decoding uses Euclidean branch metrics
  • TCM combines convolutional coding and modulation; no bandwidth expansion because symbol rate is unchanged.
  • Set partitioning hierarchically divides the expanded constellation to maximize intra-subset Euclidean distances.
  • The coded bit(s) select the subset; uncoded bits select a point within the subset.
  • Viterbi algorithm decodes using Euclidean distance branch metrics on the trellis.
  • Asymptotic coding gain for 4-state 8-PSK TCM over uncoded QPSK is approximately 3 dB.

Quick Revision

  • TCM = joint coding + modulation; key innovation by Ungerboeck (1982).
  • Bandwidth unchanged: k bits/symbol input, constellation size doubles from 2^k to 2^(k+1).
  • Set partitioning: successive division of constellation maximizes minimum intra-subset Euclidean distance.
  • Decoding: Viterbi algorithm with Euclidean distance (not Hamming) as metric.
  • ACG formula: 10 log10 (d_free_squared / d_ref_squared) dB; 4-state gives 3 dB, 8-state gives 3.98 dB.
  • Exam trap: Do NOT say TCM increases bandwidth; it only expands the constellation, not the symbol rate.
  • Applications: V.32/V.34 modems, satellite modems, some wireless standards.

TCM Concepts Quiz

Test your grasp of Trellis Coded Modulation and Ungerboeck design principles.

Question 1 of 3

Q1.In Trellis Coded Modulation, coding gain is achieved without bandwidth expansion primarily because of which design choice?