How Does Probabilistic Constellation Shaping Control Ranging Sidelobe Levels in ISAC?

Please refer to Appendix ?.

Please refer to Appendix ?.

Please refer to Appendix ?.

Please refer to Appendix ?.

Please refer to Appendix ?.

Please refer to Appendix ?.

Please refer to Appendix ?.

First note that types themselves are permutation invariant, thereby (1) would imply 2). Conversely, if 2) holds, for any realizable type q, we may choose arbitrarily a representative codeword \bm{s} satisfying \hat{P}_{\bm{s}}=q, and define h_{ \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi }(q)=J_{ \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi }(\bm{s}). If we choose another representative codeword \bm{t} in the type class { \ifcat\noexpandT\relax\mathnormal{T}\else\cal{T}\fi }_N(q), there exists a permutation matrix \ifcat\noexpand\Pi\relax\bm{\csname var\expandafter\@gobble\string\Pi\endcsname}\else\bm{\Pi}\fi such that \bm{t}= \ifcat\noexpand\Pi\relax\bm{\csname var\expandafter\@gobble\string\Pi\endcsname}\else\bm{\Pi}\fi \bm{s} since all sequences in the same type class differ only by their order of entries. Thus 2) implies h_{ \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi }(\bm{t})=h_{ \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi }(\bm{s}) for all \bm{t} and \bm{s} in the same type class, which in turn implies 1).

Let \bm{u}_n denote the n-th column of \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi. The unitarity and \ell_4 isometry imply

In addition, we have

Combining (Proof) and (Proof) yields \sum_{i<j}|u_{in}|^2|u_{jn}|^2=0, which implies that each column has at most one non-zero entry. Using again the unitarity, we see that \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi has to be monomial, and we are done.

We examine the two conditions regarding the orbitals of lines separately. Condition (a) Without loss of generality, assume that { \ifcat\noexpandL\relax\mathnormal{L}\else\cal{L}\fi }_0,\dotsc,{ \ifcat\noexpandL\relax\mathnormal{L}\else\cal{L}\fi }_{N-1} constitute an orbital. On each line { \ifcat\noexpandL\relax\mathnormal{L}\else\cal{L}\fi }_k, we choose a unit vector \bm{v}_k. Since the lines are on the same orbital, the symmetric group acts transitively on these lines, namely,

for any j, k. Therefore, for any \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \in\mathscr{S}_N, there exists a complex scalar |\lambda_{ \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi ,k}|=1 such that \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \bm{v}_k=\lambda_{ \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi ,k}\bm{v}_{\sigma_{ \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi }(k)}. On the other hand, being a permutation matrix, any \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \in\mathscr{S}_N satisfies \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \bm{1}=\bm{1}. The unit vector \bm{u}=\frac{1}{\sqrt{N}}\bm{1} thus satisfies

Taking the magnitude from both sides yields

thereby \bm{u} has constant-modulus projections on all the unit vectors \{\bm{v}_k\}. By adjusting the phases of \bm{v}_k’s we obtain

Since \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \bm{u}=\bm{u}, it further holds that \lambda_{ \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi ,k}=1. Therefore the actions of \mathscr{S}_N elements on the lines constitute a natural permutation representation \tau:\mathscr{S}_N\rightarrow \mathscr{S}_N, which is an automorphism. For N\neq 6, it is known that all automorphisms of \mathscr{S}_N are inner automorphisms [permgroup, Sec. 8.2], which implies that there exists a permutation matrix \ifcat\noexpand\Xi\relax\bm{\csname var\expandafter\@gobble\string\Xi\endcsname}\else\bm{\Xi}\fi such that

Now, let \ifcat\noexpandB\relax\bm{\csname var\expandafter\@gobble\stringB\endcsname}\else\bm{B}\fi =[\bm{v}_0,\dotsc,\bm{v}_{N-1}], we see that there exists a unitary diagonal matrix \ifcat\noexpandD\relax\bm{\csname var\expandafter\@gobble\stringD\endcsname}\else\bm{D}\fi such that \ifcat\noexpandB\relax\bm{\csname var\expandafter\@gobble\stringB\endcsname}\else\bm{B}\fi = \ifcat\noexpandV\relax\bm{\csname var\expandafter\@gobble\stringV\endcsname}\else\bm{V}\fi ^{\rm H} \ifcat\noexpandD\relax\bm{\csname var\expandafter\@gobble\stringD\endcsname}\else\bm{D}\fi, since \bm{v}_k\in \ifcat\noexpandV\relax\bm{\csname var\expandafter\@gobble\stringV\endcsname}\else\bm{V}\fi ^{\rm H}{\rm span}(\bm{e}_k). Choosing an appropriate \sigma such that \tau( \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi )\bm{e}_k=\bm{e}_{\sigma_{ \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi }(k)}, we have

and hence \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \ifcat\noexpandB\relax\bm{\csname var\expandafter\@gobble\stringB\endcsname}\else\bm{B}\fi = \ifcat\noexpandB\relax\bm{\csname var\expandafter\@gobble\stringB\endcsname}\else\bm{B}\fi \tau( \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi ). Using (Proof), we obtain

This implies that \ifcat\noexpandB\relax\bm{\csname var\expandafter\@gobble\stringB\endcsname}\else\bm{B}\fi \ifcat\noexpand\Xi\relax\bm{\csname var\expandafter\@gobble\string\Xi\endcsname}\else\bm{\Xi}\fi is a centralizer of \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \in\mathscr{S}_N. The following lemma gives a characterization of such centralizers.

Condition (b) In this case, the fact that the orbital contains N-1 lines implies that there is a homomorphism \varphi:\mathscr{S}_N\rightarrow\mathscr{S}_{N-1}. When N\geqslant 5, \mathscr{S}_N has a non-trivial normal subgroup, the alternating group \mathscr{A}_N. Since \varphi is not injective, we have \mathscr{A}_N\subseteq {\rm Ker}~\varphi for N\geqslant 5. Thus using the basic theorem of homomorphisms, we have

However, in Condition (b) the orbital size argument yields |{\rm Im}~\varphi|\geqslant N-1, which is clearly impossible for N\geqslant 5. Therefore, orbitals satisfying Condition (b) do not exist when N\geqslant 5.

The full proof is thus completed by combining the analysis of Conditions (a) and (b).

Using the classical orbital basis result on the centralizer algebra of permutation representations [cameron_permgroup, Sec. 2.2], we obtain

Using the unitarity of \ifcat\noexpandC\relax\bm{\csname var\expandafter\@gobble\stringC\endcsname}\else\bm{C}\fi, it follows that the eigenvalues on {\rm span}(\bm{1}) and its orthogonal complement are both unimodular, thus leading to (Proof).

The sufficiency is straightforward since the sidelobes of a sequence having a single non-zero entry are obviously all zero. Conversely, assume by contradiction that \bm{s} has at least two non-zero entries, let i<j be the smallest and the largest non-zero indices. We see that for k=j-i, the autocorrelation r_k=\sum_{n} s_n^*s_{n+k} is non-zero since the product s_n^*s_{n+k} is only non-zero when n=i. This implies that A_{\rm a}(\bm{s})>0, and hence the original assumption is false. This completes the proof.

Note that

Preservation of A_{\rm a} is thus equivalent to the preservation of \|\bm{s}\|_4 since \|\bm{s}\|_2^4 is already preserved due to unitarity. Now, substituting (?) into (Proof), we obtain

Using 0+n=1+(n-1), we further obtain

which further implies that

Since \sigma(k) is a bijection, a is a non-zero integer. Thus (Proof) implies (N-1)|a|\le N-1, thereby a=\pm 1. Thus we have

which corresponds to the identity and the reversal permutation, respectively. Thus the proof is completed.

First assume that the permutation component is an identity transformation. Using again the identity 0+n=1+(n-1) we have the monomial s_0s_ns_1^*s_{n-1}^*. Note that all four permitted permutations (0,n,1,n-1),~(n,0,1,n-1),~(0,n,n-1,1),~(n,0,n-1,1) correspond to the same monomial s_0s_ns_1^*s_{n-1}^*, and hence we have 4\lambda_0\lambda_n\lambda_1^*\lambda_{n-1}^*=4. This implies

When the permutation component is the reversal permutation, we may simply alter the indices accordingly. The proof is thus completed.

Let

which is an Abelian normal subgroup of { \ifcat\noexpandG\relax\mathnormal{G}\else\cal{G}\fi }_N. Note that

where \mathscr{C}_2 is the 2-cyclic group accounting for the potential reversal. If (?) holds, the commutator subgroup of \mathscr{S}_N, the alternating group \mathscr{A}_N, would have to be mapped into \widetilde{{ \ifcat\noexpandG\relax\mathnormal{G}\else\cal{G}\fi }}_N by the conjugate mapping g: \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \mapsto \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi \ifcat\noexpandG\relax\bm{\csname var\expandafter\@gobble\stringG\endcsname}\else\bm{G}\fi \ifcat\noexpandU\relax\bm{\csname var\expandafter\@gobble\stringU\endcsname}\else\bm{U}\fi ^{\rm H}. However, when N\geqslant 4, \mathscr{A}_N is non-Abelian and hence cannot be mapped into any Abelian group via the injection g. Thus we are done.