Digital Communications

Concatenation

Q: Why does a concatenated code place Reed-Solomon outside and convolutional inside?

The inner convolutional code with Viterbi decoding is excellent at correcting independent, random bit errors at low signal-to-noise ratio, but when it does fail it produces short bursts of errors rather than isolated bits. A Reed-Solomon code operates on symbols (groups of m bits, typically 8) and is naturally burst-tolerant: a single corrupted RS symbol counts as one symbol error regardless of how many of its bits are wrong. Placing RS on the outside lets it clean up the bursty residue from the Viterbi decoder. The classic CCSDS deep-space standard pairs an outer RS(255,223) code over GF(256) with an inner rate-1/2, constraint-length-7 convolutional code, with a byte interleaver of depth 2 to 8 between them to spread Viterbi error bursts across multiple RS codewords.

Q: How much coding gain does a concatenated Reed-Solomon plus convolutional code provide?

The concatenated CCSDS scheme (RS(255,223) outer, rate-1/2 K=7 convolutional inner) delivers roughly 7 to 9 dB of coding gain at an output bit error rate of 1e-5, reaching that target near Eb/N0 of about 2.5 dB. The inner convolutional code alone gives about 5 dB at the same point, and the outer RS code adds the remaining 2 to 3 dB by correcting the residual bursts. Combined code rate is 0.5 times 0.874, or about 0.437, so the bandwidth expansion is roughly 2.3 times. Turbo and LDPC codes later closed much of the remaining gap to the Shannon limit, but concatenation remained the workhorse for Voyager, Galileo, and early DVB systems.

Q: Why is an interleaver required between the inner and outer codes?

A Viterbi decoder fails in bursts: when it picks the wrong survivor path it emits a run of consecutive byte errors, often several to a few tens of bytes long. A Reed-Solomon code can only correct up to t = (n-k)/2 symbol errors per codeword; for RS(255,223) that is 16 symbols. If an entire Viterbi burst landed in one RS codeword it could exceed t and cause an uncorrectable block. A symbol interleaver of depth I reorders bytes so that consecutive Viterbi output bytes are spread across I different RS codewords, dividing any burst by I before RS decoding. CCSDS uses interleaving depth 2 to 8; depth 5 is common, ensuring bursts up to about 80 bytes are survivable.

/kon-kat-uh-NAY-shun/

Cascading two error-correcting codes in series is the core idea behind concatenation: an outer block code, classically Reed-Solomon, wraps around an inner code (usually a Viterbi-decoded convolutional code), with an interleaver between them. The inner code knocks down random bit errors at low signal-to-noise ratio, and because Viterbi failures emerge as short bursts, the symbol-oriented outer code cleans up the residue. The result is a form of forward error correction that reaches output bit error rates near 1×10^-10 at modest E_b/N₀. The CCSDS deep-space standard, RS(255,223) over a rate-1/2 constraint-length-7 convolutional code, powered the Voyager, Galileo, and Cassini downlinks and the first generation of DVB-S broadcast.

Category: Digital Communications

Typical Coding Gain: 7 to 9 dB

CCSDS Code Rate: ≈ 0.437

How Serial Code Cascading Defeats Burst Errors

The concatenated-code construction was introduced by G. David Forney in his 1966 doctoral thesis as a way to build very long, powerful codes that remain practical to decode. Rather than designing a single enormous code with an intractable decoder, concatenation splits the work between two simpler codes operating on different alphabets. The inner code works on individual bits or short symbols and is decoded with a maximum-likelihood Viterbi algorithm; the outer code works on larger symbols (bytes) over a Galois field and is decoded algebraically. Because each stage is independently decodable, the combined system inherits a minimum distance close to the product of the two component distances while keeping decoder complexity additive rather than multiplicative.

The division of labor is what makes the pairing so effective. A rate-1/2, constraint-length-7 convolutional code decoded by a soft-decision Viterbi processor gives roughly 5 dB of gain on its own, but its output errors are correlated: when the trellis search chooses the wrong survivor path it produces a run of several to a few dozen wrong bytes. A Reed-Solomon outer code is indifferent to how many bits inside a symbol are corrupted, so it treats each Viterbi burst as a handful of symbol errors. RS(255,223) corrects up to t = 16 symbol errors per 255-byte codeword, which is more than enough to absorb the residual bursts once the interleaver has spread them out.

Modern systems often replace the classic pairing with iterative schemes. Turbo codes use parallel concatenation of two recursive convolutional codes joined by an interleaver and decoded iteratively, and LDPC codes approach the Shannon limit with belief-propagation decoding. DVB-S2 retains the concatenated philosophy by pairing an outer BCH code with an inner LDPC code. The serial Reed-Solomon plus convolutional arrangement nevertheless remains in service wherever a proven, low-complexity, standards-frozen link is preferred over the last fraction of a decibel.

Code Rate and Coding Gain

Overall code rate (serial concatenation):
R_cc = R_outer × R_inner = (k/n)_out × (k/n)_in

CCSDS example:
R_cc = (223 / 255) × (1 / 2) ≈ 0.874 × 0.5 ≈ 0.437

Reed-Solomon correction capacity:
t = (n − k) / 2 = (255 − 223) / 2 = 16 symbols per codeword

Asymptotic coding gain:
G ≈ 10·log₁₀(R_cc × d_min / 2) dB

Where R = code rate, n = block length, k = information length, d_min = minimum Hamming distance. The interleaver of depth I divides any Viterbi burst length L into ceil(L / I) errors per RS codeword.

Concatenated Coding Standards

System	Outer Code	Inner Code	Interleaver	Overall Rate	BER Target
CCSDS deep space	RS(255,223)	Conv. rate-1/2, K=7	Byte, depth 2 to 8	≈ 0.437	1×10^-10
DVB-S (1994)	RS(204,188)	Conv. rate-1/2 to 7/8	Convolutional, I=12	0.41 to 0.81	1×10^-11 (QEF)
DVB-S2	BCH (t = 8 to 12)	LDPC, rate 1/4 to 9/10	Bit, block	0.25 to 0.90	1×10^-7
Turbo (3GPP)	(parallel concat.)	2 × RSC, K=4	Internal, QPP	1/3 base	1×10^-6
GSM GMSK	Block (Fire/CRC)	Conv. rate-1/2, K=5	Block, diagonal	≈ 0.45	1×10^-3

Common Questions

Frequently Asked Questions

Why does a concatenated code place Reed-Solomon outside and convolutional inside?

The inner Viterbi-decoded convolutional code excels at random bit errors but fails in short bursts. Reed-Solomon works on m-bit symbols (usually bytes) and counts one corrupted symbol regardless of how many of its bits are wrong, so placing it outside lets it absorb the bursty residue. The CCSDS standard pairs outer RS(255,223) over GF(256) with an inner rate-1/2, K=7 convolutional code and a depth-2-to-8 byte interleaver between them.

How much coding gain does a concatenated Reed-Solomon plus convolutional code provide?

The CCSDS scheme delivers roughly 7 to 9 dB of gain at an output BER of 1×10^-5, hitting that target near E_b/N₀ of about 2.5 dB. The inner code alone contributes about 5 dB; the outer RS code adds 2 to 3 dB by clearing residual bursts. Combined rate is 0.5 × 0.874 ≈ 0.437, so bandwidth expansion is about 2.3×.

Why is an interleaver required between the inner and outer codes?

A Viterbi decoder emits runs of consecutive byte errors. RS(255,223) corrects only t = 16 symbols per codeword, so an entire burst landing in one codeword could be uncorrectable. A symbol interleaver of depth I spreads consecutive Viterbi bytes across I different RS codewords, dividing each burst by I. CCSDS uses depth 2 to 8; depth 5 keeps bursts up to about 80 bytes survivable.

Related Terms

← COMSOL Concentricity →