Bounds of fast decodability of space time block codes, skew-Hermitian matrices, and Azumaya algebras

Abstract: We study fast lattice decodability of space-time block codes for $n$ transmit and receive antennas, written very generally as a linear combination $\sum_{i=1}^{2l} s_i A_i$, where the $s_i$ are real information symbols and the $A_i$ are $n\times n$ $\mathbb R$-linearly independent complex valued matrices. We show that the mutual orthogonality condition $A_iA_j^* + A_jA_i^*=0$ for distinct basis matrices is not only sufficient but also necessary for fast decodability. We build on this to show that for full-rate ($l = n^2$) transmission, the decoding complexity can be no better than $|S|^{n^2+1}$, where $|S|$ is the size of the effective real signal constellation. We also show that for full-rate transmission, $g$-group decodability, as defined in [1], is impossible for any $g \ge 2$. We then use the theory of Azumaya algebras to derive bounds on the maximum number of groups into which the basis matrices can be partitioned so that the matrices in different groups are mutually orthogonal---a key measure of fast decodability. We show that in general, this maximum number is of the order of only the $2$-adic value of $n$. In the case where the matrices $A_i$ arise from a division algebra, which is most desirable for diversity, we show that the maximum number of groups is only $4$. As a result, the decoding complexity for this case is no better than $|S|^{\lceil l/2 \rceil}$ for any rate $l$.