Summary

Chapter 9 Summary: Detection and Estimation Theory

Key Points

• 1.

  Hypothesis testing provides the rigorous framework for digital detection. The MAP rule minimises the total error probability by incorporating prior probabilities into the likelihood ratio test, while the ML rule (MAP with uniform priors) reduces to the minimum-distance detector in AWGN. The union bound $P_e \leq \sum_{m \neq \hat{m}} P(\mathbf{s}_m \to \mathbf{s}_{\hat{m}})$ provides a tractable upper bound for M-ary detection.

• 2.

  Error probability in AWGN is universally expressed through the Q-function. BPSK/QPSK achieve $P_e = Q(\sqrt{2E_b/N_0})$, and higher-order QAM/PSK follow similar closed forms. Craig's formula $Q(x) = \frac{1}{\pi}\int_0^{\pi/2} \exp\!\bigl(-\frac{x^2}{2\sin^2\phi}\bigr)\, d\phi$ is essential for averaging BER over fading distributions because the SNR appears only in the exponent.

• 3.

  Estimation theory quantifies the fundamental limits on parameter estimation. The Cramer-Rao lower bound $\operatorname{Var}(\hat{\theta}) \geq 1/I(\theta)$ sets a floor on estimator variance, with ML estimation achieving this bound asymptotically. Bayesian MMSE estimation minimises mean-square error by exploiting prior knowledge, and the LMMSE estimator provides a practical linear solution for Gaussian models.

• 4.

  Detection in fading fundamentally changes BER behaviour: the exponential decay $e^{-c \cdot \text{SNR}}$ of AWGN becomes an algebraic decay $\bar{\gamma}^{-d}$ in fading, where $d$ is the diversity order. The MGF approach provides a unified method for computing average BER across Rayleigh, Ricean, and Nakagami fading by expressing the Q-function via Craig's formula and integrating over the fading distribution analytically.

• 5.

  Channel estimation is indispensable for coherent detection. The LS estimator $\hat{\mathbf{h}}_{\text{LS}} = (\mathbf{X}^H \mathbf{X})^{-1} \mathbf{X}^H \mathbf{y}$ is simple but noisy, while the MMSE estimator $\hat{\mathbf{h}}_{\text{MMSE}} = \mathbf{R}_{hh}(\mathbf{R}_{hh} + \sigma^2(\mathbf{X}^H\mathbf{X})^{-1})^{-1}\hat{\mathbf{h}}_{\text{LS}}$ exploits channel statistics for significantly lower MSE, especially at low SNR. Pilot density must be matched to the channel's coherence bandwidth and coherence time to satisfy the Nyquist sampling theorem in frequency and time.

• 6.

  The bridge from theory to practice: detection theory tells us what performance is achievable (CRLB, optimal detector), while estimation theory tells us how accurately we can learn the channel. Imperfect CSI degrades detection: the effective SNR loss is approximately $\sigma_e^2 / (1 + \sigma_e^2)$ where $\sigma_e^2$ is the estimation error variance, motivating joint estimation-detection approaches in modern receivers.