CommLab_Sp17_Lecture_6_v0
CommLab_Sp17_Lecture_6_v0
Agenda
• Convolutional Encoding
Recap Last time we showed that ”random” linear block code combined with simple modulation (binary PAM). Can
achieve rate-efficient and energy-efficient reliable communication.
• b·G=c
• b ∈ {0, 1}1×k
• c ∈ {0, 1}1×n
• k≤n
k
• rate R = n
• In convolutional encoding the Linear Transform is a ”linear filter” (c.t. In Linear block code the Linear trnasform
is a ”matrix multiplication”)
Example 1
b = (b1 , b2 , ..., bn , ...)
1
Communication Systems Lab Spring 2017 National Taiwan University
1
Rate of this linear encoding R = 2
• In the following, we discuss how the 2 bits stored in the 2 shift registers change with the input bit stream.
observation: length of the FIR filters = number of shift registers + 1
A running example:
2
Communication Systems Lab Spring 2017 National Taiwan University
• Trellis Representation:
Generalization
∑k
• c(l) = i=1 b(i) ∗ g (i,k)
3
Communication Systems Lab Spring 2017 National Taiwan University
• Termination: In practice, due to packetization, we have to terminate at a given time K0 ≫ L (L: number of shift
registers, called ”constrained length”. When termination happens, one needs to ”reset” the registers by appending
L 0’s at the end. Rate = K0
2(K0 +L) ≈ 1
2 when K0 ≫ L
In the previous lecture, we already know that ML = MD where the distance metrics are Euclidean distance in
soft-decision, Hamming distance in hard-decision
thus
arg minb ∥ y − x(b) ∥2
b̂ =
arg minb dn (d, c(b))
• Key observation
The target function can be decomposed with respect to time
∑n
– 1. For soft-decision, the target function is ∥ y − x(b) ∥22 = i=1 |yi − xi (b)|2
∑n
– 2. For hard-decision, the target function is dH (d, c(b)) = i=1 ⊮{di ̸= ci (b)} where ⊮{·}: indicator function.
Unify the above two as a minimun cost problem b̂ = arg minAll length−n path p on the trellis {Cn (p, s)}
• Viteribi Algorithm (Dynamic Programming for solving minimun cost problem on a trellis)
let Vk (s) ≜ minAll length−k path p on the trellis {Ck (p, s)}