Ch. 6 — Performance of Digital Modulation over Wireless Channels

Chapter 6 of Wireless Communications (2nd ed. Draft) — Andrea Goldsmith

Abstract

This chapter establishes the theoretical framework for analyzing digital modulation performance over wireless channels characterized by additive white Gaussian noise (AWGN) and various fading conditions. It quantifies the trade-offs between constellation geometry, signal-to-noise ratio (SNR), and error probability metrics such as bit error probability ($P_b$) and outage probability ($P_{\text{out}}$). Central to this analysis is the derivation of average error rates using the Moment Generating Function (MGF) approach and the characterization of irreducible error floors caused by Doppler spread and intersymbol interference. These models are essential for system design, as they demonstrate how fading degrades performance from exponential decay in AWGN to linear decay in dB, necessitating diversity techniques discussed subsequently.

Key Concepts

• SNR Metrics and Energy Definitions: The performance of digital communication systems is parameterized by the received signal-to-noise power ratio, often normalized by energy metrics. Specifically, ${\gamma }_s=E_s/N_0$ represents the SNR per symbol, and ${\gamma }_b=E_b/N_0$ represents the SNR per bit, where $E_s$ and $E_b$ are the average energy per symbol and bit respectively. These quantities allow for a standardized comparison of modulation schemes independent of bandwidth or data rate, relating directly to the constellation geometry and noise PSD $N_0/2$.
• Coherent vs. Differential Detection: Coherent detection requires perfect estimation of carrier phase and frequency, yielding optimal performance in AWGN but suffering from irreducible errors in fast fading if estimation is imperfect. Differential modulation (e.g., DPSK, DQPSK) removes the need for explicit carrier recovery by detecting phase differences between consecutive symbols, incurring a $\approx 3$ dB SNR penalty in AWGN but remaining robust to slow phase drifts.
• Craig’s Q-Function Representation: To facilitate averaging error probabilities over fading distributions, the standard Gaussian Q-function is replaced by Craig’s alternate integral representation. This form, $Q(x)=\frac{1}{\pi }{\int }_0^{\pi /2}\exp \left(-\frac{x^2}{2{\sin }^2\varphi }\right)d\varphi$, allows the integration over fading SNR distributions to be performed analytically using the MGF, simplifying the performance analysis of coherent modulations like MPSK and MQAM.
• Fading Performance Metrics: In fading environments, instantaneous SNR $\gamma$ is a random variable, requiring two main performance criteria: outage probability $P_{\text{out}}$, which measures the likelihood that SNR falls below a threshold ${\gamma }_0$, and average error probability ${\bar{P}}_s$, which integrates AWGN error rates over the PDF of $\gamma$. The choice between metrics depends on the fading coherence time relative to the symbol time; slow fading necessitates outage specifications to prevent large error bursts, while fast fading allows for average error analysis.
• Moment Generating Function (MGF) Approach: The MGF of the SNR distribution, $M_{\gamma }(s)=E[e^{s\gamma }]$, serves as a unified tool for calculating average error probability in diverse fading channels (Rayleigh, Rician, Nakagami-m). By expressing the AWGN error probability as an exponential function or finite integral of such a function, the average error probability becomes a direct evaluation of the MGF at specific arguments, avoiding complex convolution integrals.
• Doppler Spread and Error Floors: High mobility introduces Doppler shift $f_D$, causing the channel phase to decorrelate over time. In differential modulation, this decorrelation creates an irreducible error floor because the reference symbol becomes a noisy estimate of the current phase. The floor magnitude depends on the product $f_D{T}_s$ (normalized Doppler spread), improving only if the data rate increases to reduce symbol time $T_s$.
• Intersymbol Interference (ISI) Limits: Frequency-selective fading induces ISI, where multipath delays cause symbols to interfere with one another. Unlike thermal noise, ISI power scales with signal power, creating an irreducible error floor dependent on the normalized RMS delay spread ${\sigma }_{{\tau }_m}/T_s$. This imposes a hard constraint on the maximum achievable data rate for a target BER without equalization.
• Composite Fading Models: Realistic channels often combine fast multipath fading (e.g., Rayleigh) with slow shadowing (log-normal). Performance analysis in such environments requires calculating the outage probability based on the shadowing component and then averaging the error probability over the fast fading conditional on the shadowed SNR. This combination ensures reliability specifications are met over spatial locations and time.

Key Equations and Algorithms

• SNR and Energy Relation: ${\gamma }_b=\frac{E_b}{N_0}=\frac{P_r{T}_b}{N_0}=\frac{P_r}{N_{0}B{\log }_{2}M}$ links received power, bandwidth, and modulation order to SNR. This equation defines the fundamental energy cost of transmitting information in the presence of noise spectral density $N_0$, establishing the baseline for comparing modulation efficiency.
• BPSK Coherent Error Probability: $P_b({\gamma }_b)=Q(\sqrt{2{\gamma }_b})$ determines the bit error rate for Binary Phase Shift Keying under coherent detection. It illustrates the exponential decay of errors with increasing SNR in an AWGN channel, serving as the benchmark for all other modulation schemes.
• Rayleigh Fading Average BER: ${\bar{P}}_b=\frac{1}{2}\left(1-\sqrt{\frac{{\bar{\gamma }}_b}{1+{\bar{\gamma }}_b}}\right)$ computes the average bit error probability for BPSK in Rayleigh fading. It demonstrates the linear decay of error probability with SNR (in linear scale) at high ${\bar{\gamma }}_b$, contrasting sharply with the exponential decay seen in AWGN.
• Craig’s Q-Function Identity: $Q(\sqrt{\gamma })=\frac{1}{\pi }{\int }_0^{\pi /2}\exp \left(-\frac{\gamma }{2{\sin }^2\varphi }\right)d\varphi$ transforms the error integral into a form amenable to averaging. This identity is critical for deriving closed-form expressions for average error probabilities over fading distributions by substituting the SNR $\gamma$ within the integral.
• MGF Method for Average Error: ${\bar{P}}_s={\int }_0^{\infty }{P}_s(\gamma )p_{\gamma }(\gamma )d\gamma$ simplifies to expressions involving $M_{\gamma }(s)$ when $P_s(\gamma )\approx \alpha Q(\sqrt{\beta \gamma })$. This algorithmic approach allows system designers to compute ${\bar{P}}_s$ by evaluating the known MGF of the fading distribution (e.g., Rayleigh, Nakagami) at specific points determined by the modulation parameters $\alpha$ and $\beta$.
• Doppler Irreducible Error Floor: $P_{\text{floor}}\approx 0.5(\pi f_D{T}_b{)}^2$ for DPSK in Rayleigh fading with uniform scattering. This approximation quantifies the performance limit imposed by channel decorrelation, showing that the error floor depends quadratically on the normalized Doppler spread and requires faster data rates to maintain a specific BER threshold.
• ISI Error Floor Bound: $P_{\text{floor}}\le K({\sigma }_{{\tau }_m}/T_s{)}^2$ bounds the bit error probability due to intersymbol interference. The constant $K$ varies by modulation (e.g., $K\approx 0.4$ for BPSK), indicating that reducing the normalized delay spread is the primary method for mitigating ISI-induced errors in frequency-selective channels.

Key Claims and Findings

• Fading Degradation from Exponential to Linear Decay: In AWGN channels, error probability decreases exponentially with SNR, whereas in Rayleigh fading, the average error probability decreases only linearly with SNR (1/ $\bar{\gamma }$) at high SNR. This fundamental shift necessitates significantly higher transmit power or diversity techniques to maintain low error rates in wireless environments compared to wired channels.
• Differential Modulation Power Penalty: Differential modulation schemes (DPSK, DQPSK) incur an approximate 3 dB SNR penalty compared to their coherent counterparts in AWGN conditions to achieve the same error probability. This penalty arises because noise adds to both the current and reference symbols in differential detection, effectively doubling the noise variance in the decision variable.
• Irreducible Error Floors in Fast Fading: Differential modulation over fast fading channels introduces an irreducible error floor that does not decrease with increasing SNR, as the limiting factor becomes phase decorrelation rather than thermal noise. The floor level is determined by the ratio of symbol time to coherence time; for a fixed fading rate, increasing the data rate lowers the floor.
• ISI Dictates Maximum Data Rate: Frequency-selective fading creates an error floor dependent on the ratio of RMS delay spread to symbol time, strictly limiting the data rate for a target BER without equalization. For example, maintaining $P_b<10^{-3}$ in a typical urban environment (${\sigma }_{{\tau }_m}\approx 2.5\mu s$) restricts the data rate to approximately 40 kbaud for BPSK unless diversity or equalization is employed.
• MGF Unifies Fading Analysis: The Moment Generating Function approach provides a unified framework for calculating average error probabilities across different fading distributions (Rayleigh, Rician, Nakagami) by expressing AWGN error rates as integrals of the noise exponential. This eliminates the need for separate numerical integrations for each fading model, enabling rapid system performance evaluation.
• Outage Probability for Slow Fading: When fading coherence time is much longer than the symbol time, average error probability is an insufficient metric; instead, outage probability must be used to ensure service availability for a specified percentage of time or locations. The required fade margin increases logarithmically with the stringency of the outage requirement (e.g., $20$ dB for 0.01% outage).

Terminology

• Outage Probability ($P_{\text{out}}$): The probability that the instantaneous SNR $\gamma$ falls below a specified threshold ${\gamma }_0$ required for acceptable quality of service. It is used primarily in slow fading scenarios where deep fades persist long enough to cause unrecoverable burst errors.
• Average Error Probability (${\bar{P}}_s$): The expected value of the symbol error probability averaged over the probability distribution of the fading SNR. This metric is appropriate when the channel varies rapidly enough (fast fading) that individual fades do not dominate the long-term performance.
• Moment Generating Function ($M_{\gamma }(s)$): Defined as $E[e^{s\gamma }]$, this statistical tool characterizes the distribution of the random SNR variable $\gamma$. It is used to compute average error rates over fading channels by integrating the error probability expression against the fading PDF.
• Doppler Spread ($f_D$): The spectral broadening of the received signal caused by relative motion between transmitter and receiver, equal to $v/\lambda$. It determines the channel’s coherence time and dictates the severity of phase decorrelation for differential modulation.
• Coherence Time ($T_c$): The time duration over which the channel impulse response is essentially invariant, inversely proportional to the Doppler spread ($T_c\approx 1/f_D$). If the symbol time $T_s>T_c$, the channel is considered slowly fading for that transmission.
• Irreducible Error Floor: A lower bound on the achievable error probability that cannot be reduced by increasing transmit power. It arises in differential modulation due to Doppler-induced phase decorrelation or in frequency-selective channels due to intersymbol interference.
• Normalized RMS Delay Spread (${\sigma }_{{\tau }_m}/T_s$): The ratio of the channel’s RMS delay spread to the symbol time. This dimensionless parameter governs the severity of intersymbol interference; values significantly greater than 0.1 typically necessitate equalization to avoid high error floors.
• Fade Margin ($F_d$): The additional SNR (in dB) required above the minimum operating threshold to ensure a specific outage probability is met in a fading environment. It accounts for the statistical variations in received power due to shadowing and multipath fading.
• Gray Encoding: A bit-to-symbol mapping technique where adjacent constellation points differ by exactly one bit position. This minimizes the bit error probability $P_b$ relative to the symbol error probability $P_s$ in high-SNR regimes, allowing the approximation $P_b\approx P_s/{\log }_{2}M$.
• Diversity Order ($M$): The exponent $k$ in the asymptotic approximation of the average error rate ${\bar{P}}_s\propto {\bar{\gamma }}^{-k}$. In a system with $M$ independent fading paths combined optimally, the diversity order is $M$, indicating the slope of the error rate curve in the high-SNR regime.