Ch. 5 — Digital Modulation and Detection

Chapter 5 of Wireless Communications (2nd ed. Draft) — Andrea Goldsmith

Abstract

Chapter 5 establishes the fundamental framework for digital modulation and detection in wireless communication systems, transitioning from information-theoretic capacity limits to practical signal design. The central technical contribution is the signal space analysis method, which maps time-domain waveforms to finite-dimensional vectors to facilitate optimal receiver design and error probability calculation. This chapter details linear and nonlinear modulation schemes, pulse shaping techniques, and synchronization requirements, providing the necessary tools to optimize the trade-off between spectral efficiency, power efficiency, and implementation complexity.

Key Concepts

• Signal Space Analysis: Signals are represented as vectors in a finite-dimensional vector space spanned by $N$ orthonormal basis functions ${\varphi }_j(t)$. This geometric transformation converts infinite-dimensional function spaces into ${\mathbb{R}}^N$, simplifying the analysis of distance and error probability using vector operations. It applies to any digital modulation scheme where bits are mapped to analog signals $s_i(t)$.
• Optimal Detection and Sufficient Statistics: The receiver minimizes message error probability by maximizing the posterior probability $p(s_i\mid r)$ given the received signal $r(t)$. The projection $r_j={\int }_0^{T}r(t){\varphi }_j(t)dt$ forms a sufficient statistic vector $\mathbf{r}$, allowing the discard of remainder noise orthogonal to the signal space without performance loss. This justifies the use of matched filter receivers.
• Maximum Likelihood Decision Criterion: For equally likely messages, the optimal receiver selects the constellation point $s_i$ closest to the received vector $\mathbf{r}$ in Euclidean distance. The decision regions $Z_i$ partition the signal space such that any $\mathbf{r}\in Z_i$ maps to message $m_i$. This criterion minimizes symbol error probability under Additive White Gaussian Noise (AWGN) conditions.
• Linear Modulation (MPAM, MPSK, MQAM): Information is encoded in amplitude, phase, or both. M-ary Polar Amplitude Modulation (MPAM) uses 1-D constellations; M-ary Phase Shift Keying (MPSK) uses constant envelope 2-D constellations; and M-ary Quadrature Amplitude Modulation (MQAM) uses 2-D square constellations for higher spectral efficiency.
• Differential Modulation (DPSK/DQPSK): To avoid the complexity of coherent phase recovery, information is encoded as phase transitions relative to the previous symbol. A 0-bit corresponds to no phase change, while a 1-bit corresponds to a phase shift of $\pi$ in DPSK. This approach introduces an error floor in fading channels due to Doppler decorrelation.
• Frequency Modulation (FSK, MSK, CPFSK): Information is encoded in the carrier frequency, resulting in a constant envelope which permits efficient nonlinear amplification. Minimum-Shift Keying (MSK) represents binary FSK with minimum frequency separation $0.5/T_s$ for orthogonality. Continuous-Phase FSK (CPFSK) ensures phase continuity to reduce spectral broadening.
• Pulse Shaping and Nyquist Criterion: Baseband pulse shapes $g(t)$ are applied to limit spectral occupancy and eliminate intersymbol interference (ISI). To satisfy the Nyquist criterion, the effective pulse $p(t)$ must be zero at all sampling instants $k{T}_s$ for $k\ne 0$. Raised cosine pulses achieve this with a controllable rolloff factor $\beta$.
• Carrier Phase and Timing Synchronization: Coherent detection requires the receiver to estimate the carrier phase $\varphi$ and symbol timing $\tau$ from the received signal $r(t)$. Maximum Likelihood (ML) estimation minimizes the integral of the squared error between the received and hypothesized signals. Phase-Locked Loops (PLL) are used to track slow phase variations dynamically.
• Gray Coding: A constellation mapping technique where adjacent signal points in the geometric space differ by exactly one bit in their binary representation. This minimizes the bit error rate (BER) when the most probable symbol errors occur to nearest neighbors, approximating $P_b\approx P_e/{\log }_{2}M$.
• Constellation Shaping and Offset: Shaping constellations to approximate spheres can save up to 1.3 dB of power but increases complexity. Quadrature Offset (e.g., OQPSK) shifts the quadrature branch by $T_s/2$ to prevent simultaneous I/Q transitions, reducing amplitude fluctuations and nonlinear distortion.

Key Equations and Algorithms

• Signal Space Representation: $s_i(t)={\sum }_{j=1}^N{s}_{ij}{\varphi }_j(t)$. This decomposes the transmitted signal into projection coefficients $s_{ij}$ along orthonormal basis functions. It maps the time function to a vector ${\mathbf{s}}_i=(s_{i1},\dots ,s_{iN})$ in ${\mathbb{R}}^N$.
• Coefficient Calculation: $s_{ij}={\int }_0^T{s}_i(t){\varphi }_j(t)dt$. This integral computes the dot product of the signal and the basis function, determining the vector component. It relies on the basis orthonormality ${\int }_0^T{\varphi }_j(t){\varphi }_k(t)dt={\delta }_{jk}$.
• Euclidean Distance: $d_{ik}=\Vert {\mathbf{s}}_i-{\mathbf{s}}_k\Vert =\sqrt{{\int }_0^T|s_i(t)-s_k(t){|}^{2}dt}$. This metric defines the separation between constellation points. The minimum distance $d_{min}$ primarily dictates the system’s error performance in AWGN.
• Union Bound on Error Probability: $P_e\le {\sum }_{k\ne i}Q\left(\frac{d_{ik}}{\sqrt{2N_0}}\right)$. This upper bound approximates the symbol error probability for equally likely messages. The Q-function represents the tail probability of a standard Gaussian random variable.
• Nearest Neighbor Approximation: $P_e\approx M_{d_{min}}Q\left(\frac{d_{min}}{\sqrt{2N_0}}\right)$. At high SNR, errors are dominated by confusions with nearest neighbors. $M_{d_{min}}$ is the average number of nearest neighbors in the constellation.
• Raised Cosine Spectrum: $P(f)=\{\begin{array}{ll}T,& |f|\le \frac{1-\beta }{2T}\\ \frac{T}{2}\left[1+\cos \left(\frac{\pi T}{\beta }(|f|-\frac{1-\beta }{2T})\right)\right],& \frac{1-\beta }{2T}<|f|\le \frac{1+\beta }{2T}\\ 0,& |f|>\frac{1+\beta }{2T}\end{array}$. This frequency-domain definition ensures zero ISI. $\beta \in [0,1]$ controls bandwidth expansion; $\beta =0$ is the ideal bandlimited case.
• ML Phase Estimation Condition: $z(t)={\int }_0^{T_0}r(t)\sin (2\pi f_{c}t+\hat{\varphi })dt\approx 0$. This condition characterizes the lock point of a Phase-Locked Loop (PLL). The integrator adjusts the Voltage-Controlled Oscillator (VCO) phase $\hat{\varphi }$ until the quadrature component vanishes.
• MSK Frequency Separation: $\Delta f_c=\frac{0.5}{T_s}$. This specifies the minimum frequency separation for orthogonality in Minimum-Shift Keying. It ensures $⟨s_i(t),s_j(t)⟩=0$ for different symbols, enabling optimal coherent detection.
• GMSK Bandwidth Constraint: $B_g{T}_s=0.5$ is cited as a tolerable level for voice transmission. This parameter balances spectral efficiency against inter-symbol interference (ISI) inherent in Gaussian pulse shaping which does not satisfy the Nyquist criterion.
• Average Energy for MPAM: $E_{avg}=\frac{M^2-1}{12}4d^2{\int }_0^{T_s}{g}^2(t)dt$. This calculates the average transmit energy for an MPAM constellation with minimum distance $2d$. It depends on the constellation size $M$ and the pulse shape energy.

Key Claims and Findings

• Digital modulation strictly outperforms analog modulation in spectral efficiency, error correction potential, and security, despite typically requiring higher implementation complexity.
• Optimal detection performance in AWGN is achieved by maximizing the likelihood function, which is equivalent to finding the closest constellation point in Euclidean distance.
• Gray coding reduces the bit error probability by a factor of approximately ${\log }_{2}M$ compared to uncoded mapping for the same symbol error rate, provided minimum distance errors dominate.
• Frequency modulation schemes like MSK and GMSK offer power efficiency through constant amplitude transmission but incur a spectral efficiency penalty compared to linear modulations like QAM.
• Pulse shaping using raised-cosine filters eliminates intersymbol interference at sampling instants, enabling reliable recovery at the Nyquist rate.
• Differential modulation eliminates the need for coherent carrier recovery but suffers from an irreducible error floor in frequency-selective or rapidly fading channels due to phase decorrelation.
• The spectral efficiency of GMSK is optimized at the cost of introducing controllable ISI, which is acceptable for voice links but requires equalization for high-data-rate applications.

Terminology

• Basis Function: An element ${\varphi }_j(t)$ of an orthonormal set used to expand signals into a vector representation, satisfying $\int {\varphi }_j(t){\varphi }_k(t)dt={\delta }_{jk}$.
• Constellation Point: The vector ${\mathbf{s}}_i\in {\mathbb{R}}^N$ representing a message $m_i$ in signal space, corresponding to the transmitted signal $s_i(t)$.
• Decision Region: The subset $Z_i\subset {\mathbb{R}}^N$ of the signal space where a received vector $\mathbf{r}$ is decoded as message $m_i$, defined by the ML criterion.
• Minimum Distance ($d_{min}$): The smallest Euclidean distance between any pair of distinct constellation points, determining the primary error event in high SNR regimes.
• Sufficient Statistic: A reduced set of data $\mathbf{r}$ derived from $r(t)$ that contains all information necessary to make an optimal decision, discarding orthogonal noise components.
• Coherent Detection: A demodulation technique requiring the receiver to know the carrier phase $\varphi$ and timing $\tau$ of the transmitted signal to correctly interpret in-phase and quadrature components.
• Nyquist Criterion: A condition on the pulse shape $p(t)$ requiring $p(k{T}_s)=0$ for all integers $k\ne 0$ to prevent inter-symbol interference in the absence of channel distortion.
• Rolloff Factor ($\beta$): A parameter in raised-cosine pulses ($0\le \beta \le 1$) that determines the bandwidth excess beyond the Nyquist bandwidth $1/2T_s$ and the rate of spectral decay.
• Voltage-Controlled Oscillator (VCO): A circuit in a Phase-Locked Loop (PLL) that generates the local carrier frequency, adjusted by a control voltage derived from the phase error signal.
• GMSK: Gaussian Minimum-Shift Keying, a CPFSK variant using a Gaussian pulse shape to improve spectral efficiency, used in GSM standards but introducing ISI.