Explanation of proof of optimal decoding via factor group

28 Views Asked by At

Consider an additive map $\varphi\colon\mathbb{F}_2^m\to\mathbb{F}_2^n$ and its image $C$. (Interpret $C$ as a code)

Now, optimal decoding maps $x\in\mathbb{F}_2^n$ to the nearest $c\in C$ in terms of the hamming-distance.

Consider the factorgroup $\mathbb{F}_2^n/C$ and use as representatives of the cosets the elements of the coset of minimum hemming-weight: $$\begin{array}{rclrrcr} C &= & \{&0,& c_1,&\dots,&c_{2^m-1}\}\\ b_1+C &=& \{&b_1,& b_1+c_1,&\dots,&b_1+c_{2^m-1}\}\\ b_2+C &=& \{&b_2,& b_2+c_1,&\dots,&b_2+c_{2^m-1}\}\\ & \dots&&&&&\\ b_{2^{n-m}-1}+C &=& \{&b_{2^{n-m}-1},& b_{2^{n-m}-1}+c_1,&\dots,&b_{2^{n-m}-1}+c_{2^m-1}\}\end{array}$$

Now, for $x=b_j+c_i\in\mathbb{F}_2^n$ the nearest code-word is $c_j$, the top-element of x's columns.

I have problems understanding the proof for this theorem, that goes as follows:

Suppose $c_{j_0}$ was sent and $x=c_{j_0}+e$ arrived. Consider the coset $e+C$, and represent it as above: $e+C=b_{j_0}+C$ for some $j_0$. Now, as $x=b_{j_0}+c_{k_0}$ for some $k_0$, $c_{k_0}$ ist the nearest code-word.

Why? The steps are too big for me.

Thanks alot in advance

1

There are 1 best solutions below

0
On BEST ANSWER

The key is that we have built this standard array using

as representatives of the cosets the elements of the coset of minimum hamming-weight

(which in general is not required, BTW).

That means that we have chosen $b_i$ (coset leader) as the element with minimum weight inside the coset: $w(b_j) \le w(b_j +C_i)$

Now we receive $x = C + e $ for some unknown $C$ that we will try to guess (lets call this guess $\hat C$). And we find that $x$ is in the position $(i,j)$ in our standard array, so $x = C_i+ b_j$.

To guess that $\hat C =C_t $ for some $t$ implies to guess an error

$$\hat e= x + \hat C=C_i+b_j + C_t$$

If we choose $i=t$, we get $\hat e= b_j$

Elsewhere , because $C_i + C_t = C_s$, for some $s\ne 0$, we get $\hat e = b_j +C_s$. This corresponds to another element of the same coset.

But we have stated above that $w(b_j) \le w(b_j +C_i)$. Hence the first alternative gives us the guessed error with minimum weight.