Ch. 9 — Adaptive Modulation and Coding

Chapter 9 of Wireless Communications (2nd ed. Draft) — Andrea Goldsmith

Abstract

This chapter establishes the theoretical framework for adaptive modulation and coding (AMC) in wireless fading channels, focusing on joint optimization of transmission rate and power to maximize spectral efficiency while satisfying bit error probability constraints. It derives optimal continuous and discrete adaptation policies, demonstrating that variable-rate variable-power MQAM achieves performance within a constant spectral efficiency gap $K$ of the Shannon capacity. The chapter further analyzes practical implementation constraints, including discrete constellation sets, channel estimation errors, feedback delays, and composite fast-slow fading environments, providing design rules for robust wireless system deployment.

Key Concepts

• Variable-Rate Variable-Power MQAM: This technique adjusts both the modulation constellation size $M$ and transmit power $P$ based on instantaneous channel SNR $\gamma$ to maintain a target bit error probability $P_b$. By utilizing a water-filling-like power allocation policy, it maximizes the average spectral efficiency compared to fixed-rate schemes, particularly in high SNR regimes.
• Adaptive Coding: While this chapter focuses on adaptive modulation, it acknowledges adaptive coding as a parallel mechanism typically implemented via puncturing or multiplexing (e.g., in GSM/EDGE) to handle channel variations when modulation is fixed. Hybrid techniques combine both to optimize rate, power, coding, and error probability simultaneously.
• Water-Filling Power Adaptation: The optimal power control policy for maximizing spectral efficiency under an average power constraint $\bar{P}$. The transmitter allocates power inversely proportional to the channel fade depth $\gamma$ above a cutoff ${\gamma }_0$, effectively filling the “good” channel states while avoiding transmission during deep fades.
• Channel Inversion: A power adaptation strategy where transmit power $P(\gamma )$ is adjusted to maintain a constant received SNR, thereby converting the fading channel into an equivalent AWGN channel. While this simplifies receiver design, it suffers a significant spectral efficiency penalty compared to variable-rate techniques, particularly under Rayleigh fading.
• Discrete-Rate Adaptation: A practical implementation constraint where the constellation size $M$ is restricted to a finite set $\{M_0,\dots ,M_{N-1}\}$ rather than varying continuously with $\gamma$. This discretization requires grouping channel states into fading regions $R_j$, incurring a negligible spectral efficiency loss of approximately 1 dB with only 5 to 6 regions.
• Average Fade Region Duration (AFRD): A metric estimating the time ${\bar{\tau }}_j$ the channel SNR $\gamma$ remains within a specific fading region $R_j$ before transitioning to an adjacent region. It is crucial for determining the required rate of constellation switching and feedback loop latency to prevent adaptation errors in fast fading environments.
• Adaptive Coded Modulation: A scheme that superimposes trellis or lattice codes onto adaptive modulation constellations to recover the coding gain $G_c$ lost through uncoded transmission. The power adaptation policy remains structurally identical to the uncoded case but incorporates the coding gain into the effective power loss factor $K_c$.
• General M-ary Modulation Bounds: A unified BER approximation framework $P_b(\gamma )=c_1\exp (-c_2\gamma /(M-1))$ used to extend adaptive policies beyond MQAM to other modulations like MPSK. Depending on the sign of the constant $c_4$ in extended bounds, optimal power adaptation may resemble water-filling, inverse water-filling, or on-off switching.
• Composite Fading Adaptation: An extension of adaptive techniques to channels combining fast fading (Rayleigh) and slow fading (log-normal shadowing). The transmitter adapts only to the slow fading component $\bar{\gamma }$ because fast fading varies too rapidly for accurate feedback, optimizing rate and power based on the long-term average SNR.
• Channel Estimation Error and Delay: Practical non-idealities where the transmitter uses a noisy estimate $\hat{\gamma }$ or a delayed version $\gamma [t-d]$ of the channel state. These factors degrade BER performance, requiring the estimation error $ϵ$ to be kept within 1 dB and normalized delay $i_d{f}_D$ below 0.001 to maintain target reliability.

Key Equations and Algorithms

• MQAM BER Bound (Tight): $P_b\approx \frac{1}{4}\exp \left(-\frac{1.5\gamma }{M-1}\right)$. This expression bounds the bit error probability for MQAM with $M$ constellation points and received SNR $\gamma$, valid for $M\ge 4$. It serves as the invertible basis for deriving optimal constellation size as a function of channel quality.
• Optimal Constellation Size: $M(\gamma )=1+\frac{1.5\gamma }{-\ln (5P_b)}$. Derived by inverting the BER bound, this equation determines the maximum real-valued constellation size supportable for a given instantaneous SNR $\gamma$ and a target BER constraint $P_b$.
• Optimal Power Adaptation: $\frac{P(\gamma )}{\bar{P}}=\left(\frac{1}{{\gamma }_0}-\frac{1}{\gamma }\right)$. This water-filling policy dictates the transmit power ratio relative to the average power $\bar{P}$ for $\gamma \ge {\gamma }_0$. The cutoff depth ${\gamma }_0$ is determined by the constraint $E[P(\gamma )]=\bar{P}$.
• Average Spectral Efficiency: $\bar{R}={\int }_{{\gamma }_0}^{\infty }{\log }_2\left(1+\frac{1.5\gamma }{-\ln (5P_b)}\right)p(\gamma )d\gamma$. This integral computes the expected data rate per unit bandwidth by averaging the instantaneous rate ${\log }_{2}M(\gamma )$ over the fading distribution $p(\gamma )$, yielding the system’s capacity under the adaptation policy.
• Discrete Region Mapping: $R_j=[{\gamma }_j,{\gamma }_{j+1})$ where ${\gamma }_j={\gamma }_{K^{\ast }}{M}^{j+1}$. This algorithm defines the boundaries for discrete-rate adaptation by partitioning the SNR range into regions associated with discrete constellation sizes $M_j$, ensuring the largest feasible constellation is selected for a given $\gamma$.
• Average Fade Region Duration: ${\bar{\tau }}_j=\frac{{\pi }_j}{L({\gamma }_j)+L({\gamma }_{j+1})}$. This approximation uses the steady-state probability ${\pi }_j$ and level-crossing rates $L$ at region boundaries to estimate the duration the channel remains in the $j$-th fading region, assuming a finite-state Markov model.
• Truncated Channel Inversion Spectrum Efficiency: $\bar{R}={\log }_2\left(1+\frac{\bar{\gamma }}{E[1/\gamma ]}\right)(1-P_{out})$. This equation calculates the efficiency of maintaining a fixed rate by inverting the channel gain above a cutoff ${\gamma }_0$, weighted by the probability of outage $P_{out}=P(\gamma \le {\gamma }_0)$.
• Adaptive Coded Modulation Rate: $R_{data}={\log }_{2}M(\gamma )-\frac{2r}{N}$. This formula adjusts the data rate by subtracting the overhead required for the uncoded bits that select the coset point in a trellis-coded scheme, where $r$ is the code rate and $N$ is the number of bits per symbol.
• General M-ary BER Form: $P_b\approx c_1\exp \left(\frac{-c_2\gamma }{c_{3}M+c_4}\right)$. This generalized approximation facilitates optimization for various modulation types by capturing the dependency between BER, SNR $\gamma$, and constellation size $M$ through constants $c_i$ derived via curve fitting.
• Composite Fading Power Adaptation: $\frac{P(\bar{\gamma })}{\bar{P}}=\frac{1}{{\gamma }_0}-\frac{1}{\bar{\gamma }}$. In composite channels, the water-filling structure persists but applies to the slow fading average SNR $\bar{\gamma }$, adapting transmission resources to shadowing variations while averaged over fast fading realizations.

Key Claims and Findings

• Adaptive MQAM with variable rate and variable power achieves average spectral efficiency within a constant gap $K$ of the Shannon capacity for flat fading channels, independent of the fading distribution.
• Restricting the adaptive policy to just five or six discrete fading regions results in a spectral efficiency loss of approximately 1 dB compared to the continuous-rate optimum, making discrete implementation highly desirable.
• Truncated channel inversion with a fixed rate achieves spectral efficiency nearly identical to variable-rate variable-power MQAM in Rayleigh fading, but requires a higher outage probability ranging from 0.1 to 0.6 to maximize efficiency.
• Channel estimation error must be maintained below 1 dB (for $P_b=10^{-3}$) and feedback delay normalized by Doppler frequency ($i_d{f}_D$) must be less than 0.001 to prevent significant BER degradation.
• For general M-ary modulations like MPSK, the form of optimal power adaptation depends on the specific BER bound coefficients; certain bounds result in inverse water-filling or on-off power control rather than standard water-filling.
• In composite fading environments, adaptation based solely on slow fading shadowing $\bar{\gamma }$ yields an optimal policy that maintains the water-filling structure relative to $\bar{\gamma }$ rather than instantaneous SNR.

Terminology

• Spectral Efficiency: The total data rate $R$ normalized by bandwidth $B$, typically expressed in bits per second per Hertz (bps/Hz), representing the bandwidth efficiency of the transmission scheme.
• Water-Filling: A power allocation strategy where more transmit power is assigned to channel states with higher gain (lower fade depth) and zero power below a cutoff ${\gamma }_0$, analogous to pouring water into a vessel with variable bottom levels.
• Outage Probability: The probability $P_{out}=P(\gamma \le {\gamma }_0)$ that the channel condition falls below a specific cutoff threshold, leading to no data transmission or failure to meet the BER target.
• Level-Crossing Rate: The frequency at which the channel SNR $\gamma$ crosses a specific threshold level in the positive or negative direction, used to estimate state transition probabilities in fading models.
• Fade Region Duration: The average time ${\bar{\tau }}_j$ the channel signal quality remains within a specific region $R_j$ of the SNR range, determining the minimum latency required for state updates.
• Cutoff Fade Depth (${\gamma }_0$): The minimum received SNR threshold below which the system ceases transmission to conserve power, determined by the average power constraint and fading statistics.
• Coding Gain ($G_c$): The SNR reduction required by a coded system to achieve the same BER as an uncoded system, which in adaptive modulation acts as an effective power increase.
• Composite Fading: A propagation model combining fast multipath fading (Rayleigh) and slow shadowing (log-normal), where adaptation is feasible only for the slow fading component due to mobility constraints.
• Discrete-Rate Adaptation: An implementation of AMC where the modulation order is constrained to a discrete set $\{M_j\}$ rather than varying continuously, reducing hardware complexity and signaling overhead.
• Ergodic Capacity: The maximum achievable rate averaged over the fading distribution, achievable with adaptive coding when the delay constraint allows for coding over all channel states.