In this section, we will analyze the error performance of the cooperative system employing the DUSTC. Under high SNR assumption, the upper bound of CER,
P
e
, will be derived using the proposed scheme. The CER can be found by calculating the pairwise CER between the source and destination. Let us first consider the covariance matrix of the aggregate noise
\(\tilde{{\user2{Z}}}_n^{\prime d}\) in Eq.
6, which bears the form
\(\hbox{diag} \{\sigma^2_{h_{1,eff}},\sigma^2_{h_{2,eff}},\ldots,\sigma^2_{h_{L,eff}}\}\), where the corresponding
kth diagonal entry of the covariance matrix is given by
$$ \sigma_{h_{k,eff}}^2= 2 {\mathcal{N}}_0 ({\mathcal{E}}_{r_k}A_{r_k}^2 \sigma_{r_kd}^2 + 1), \quad k=1,2, \ldots, L. $$
(8)
To normalize the aggregate noise variance, let us define the matrix
\({\user2{G}}:= \hbox{diag}\{g_1, g_2,\ldots, g_L\}\) with
\(g_k=(\mathcal{E}_{r_k}A_{r_k}^2 \sigma_{r_kd}^2+ 1)^{-1/2}\). Then, by multiplying
\({\user2{G}}\) with the received signal block at the destination, we can rewrite Eq.
6 as
$$ {\user2{Y}}^{d}_n{\user2{G}} ={\user2{Y}}^{d}_{n-1}{\user2{V}}_n^{(m)}{\user2{G}}+\tilde{{\user2{Z}}}_n^{\prime d}{\user2{G}}, $$
(9)
or equivalently, we have
$$ \tilde{{\user2{Y}}}^{d}_n ={\user2{Y}}^{d}_{n-1}\tilde{{\user2{V}}}_n^{(m)}+\tilde{{\user2{Z}}}_n, $$
(10)
where
\(\tilde{{\user2{Y}}}^{d}_n={\user2{Y}}^{d}_n{\user2{G}}, \tilde{{\user2{V}}}^{(m)}_n={\user2{V}}^{(m)}_n{\user2{G}}\), and
\(\tilde{{\user2{Z}}}_n=\tilde{{\user2{Z}}}^{\prime d}_n{\user2{G}}\). Then, the conditional CER can be calculated using Eq.
10, which is given by
$$ \begin{aligned} P[\tilde{{\user2{V}}}_n \rightarrow \tilde{{\user2{V}}}^{\prime}_n|{\user2{Y}}_{n-1}] &= Q \left( \sqrt{ D_c^2(\tilde{{\user2{V}}}_n, \tilde{{\user2{V}}}^{\prime}_n) / 4 {\mathcal{N}}_0 } \right) \\ &\leq \exp \left[ - D_c^2(\tilde{{\user2{V}}}_n, \tilde{{\user2{V}}}^{\prime}_n) / 8 {\mathcal{N}}_0\right], \end{aligned} $$
(11)
where
\({D_c^2(\tilde{{\user2{V}}}_n, \tilde{{\user2{V}}}^{\prime}_n) =\hbox{tr}\{ {\user2{Y}}_{n-1}({\user2{V}}_ {n}-{\user2{V}}^{\prime}_{n}){\user2{G}}{\user2{G}}^{\mathcal{H}}({\user2{V}}_{n}-{\user2{V}}^{\prime}_{n})^{\mathcal{H}} {\user2{Y}}_{n-1}^{\mathcal{H}} \}}\). Using the relationship
\({\user2{Y}}^{d}_n \approx \tilde{{\user2{H}}}_n{\user2{X}}^s_n\) at high SNR (c.f. Eq.
6), the code distance can be approximated as
$$ D_c^2(\tilde{{\user2{V}}}_n, \tilde{{\user2{V}}}^{\prime}_n) \approx \hbox{tr}\{ ({\user2{H}}^{d}_{n} {\user2{H}}^{r}_{n})\Updelta_e({\user2{H}}^{d}_{n} {\user2{H}}^{r}_{n})^{{\mathcal{H}}} \}, $$
(12)
where
\({\Updelta_e= \mathcal{E}_s{\user2{E}}^{1/2}_r{\user2{X}}_{n-1} ({\user2{V}}_n-{\user2{V}}^{\prime}_n)({\user2{A}}{\user2{G}}) ({\user2{A}}{\user2{G}})^{\mathcal{H}}({\user2{V}}_n- {\user2{V}}^{\prime}_n)^{\mathcal{H}}{\user2{X}}_{n-1}^{\mathcal{H}} {\user2{E}}^{1/2}_r}\). Since
\(\Updelta_e\) is Hermitian, we can perform the following decomposition:
\({\Updelta_e= {{\user2{U}}^{\prime}}^{\mathcal{H}}{\user2{D}}_e{\user2{U}}^{\prime}}\), where
\({\user2{U}}^{\prime}\) is a unitary matrix, and
\({\user2{D}}_e\) is
\(\hbox{diag}\{\lambda_{e,1},\lambda_{e,2},\ldots,\lambda_{e,L}\}\). Each diagonal entry
\(\lambda_{e,k}, k=1,2,\ldots,L\), represents an eigenvalue of
\({\user2{D}}_e\). Then, the CER can be obtained by averaging Eq.
11 with respect to channel realizations
\({\user2{H}}^{d}_{n} {\user2{H}}^{r}_{n}\). Let us define
\(h:=h^{r_kd}h^{sr_k}\), then the probability density function (PDF) of α = |
h| is given by [
10]
$$ p_{\alpha}(\alpha)=\frac{4\alpha}{\sigma^2_{r_kd}\sigma^2_{sr_k}} K_0\left(2\sqrt{\frac{\alpha^2} {\sigma^2_{r_kd}\sigma^2_{sr_k}}}\right), $$
(13)
where
\(K_0(\cdot)\) is the zeroth order modified Bessel function of the second kind. By assuming each fading coefficient has unit variance, the CER can be computed by using the integration property of Bessel functions [
5, Eq. 6.631.3] and follows:
$$ \int_{0}^{\infty}x^\theta e^{-\alpha x^2}K_{\phi}(\beta x)dx\\ =\frac{1}{2}\alpha^{-\frac{1}{2}\theta}\beta^{-1} \Upgamma\left(\frac{1+\theta+\phi}{2}\right) \times \Upgamma\left(\frac{1-\phi+\theta}{2}\right)e^{\frac{\beta^2}{8\alpha}} W_{-\frac{\theta}{2},\frac{\phi}{2}}\left(\frac{\beta^2} {4\alpha}\right),\\ $$
(14)
where
\(W_{m,n}(z)=e^{-z/2}z^{n+1/2}U(1/2+n-m,1+2n,z)\) is the Whittaker function with
\(U(\cdot,\cdot,\cdot)\) denoting confluent hypergeometric function of the second kind. By further using the approximation
\(U(a,1,1/x)\approx \hbox{ln}(x)/\Upgamma(a)\) [
1, Eqs. 13.5.9] at high SNR, the CER can be simplified to
$$ P_{e}=P[\tilde{{\user2{V}}}_n \rightarrow \tilde{{\user2{V}}}^{\prime}_n]\leq \prod _{k=1} ^{L} \left[\ln\left(\frac{\lambda_{e,k}}{8 {\mathcal{N}}_0}\right)\right]\left(\frac{\lambda_{e,k}}{8 {\mathcal{N}}_0}\right)^{-1}. $$
(15)
It is worth mentioning that the CER of cooperative networks is very reminiscent of its counterpart, namely the multi-input single-output (MISO) system with co-located mutiple transmit antennas employing the DUSTC (see e.g., [
13]). Their difference lies in the log term which reflects the effect of the amplification and the aggregate noise. However, this log term only leads to a coding gain loss. In fact, Eq. (
15) also confirms that cooperative networks provide full diversity gain proportional to the number of relays
L.