Adaptive Delta Modulation

Adaptive Delta Modulation (ADM) is an improved version of basic Delta Modulation that removes the fixed step size constraint. By allowing the step size to vary dynamically based on the characteristics of the input signal, ADM significantly reduces both slope overload distortion and granular noise simultaneously. This makes ADM one of the most practically useful pulse modulation techniques for voice and audio encoding.

Core Concept of Adaptive Delta Modulation

In basic Delta Modulation, the fixed step size causes an inherent conflict: a small step reduces granular noise but causes slope overload for fast signals, while a large step avoids slope overload but increases granular noise for slow signals. **Adaptive Delta Modulation** resolves this by continuously adjusting the step size delta(n) based on the pattern of recent output bits.

The adaptation rule examines the transmitted bit stream. If consecutive 1s or consecutive 0s are detected, it indicates slope overload is occurring because the staircase is repeatedly stepping in the same direction trying to catch the signal. In this case, the step size is increased. If the output alternates between 1 and 0, it indicates granular noise, meaning the signal is flat and the staircase is oscillating. In this case, the step size is decreased.

The simplest ADM adaptation algorithm is the one-bit feedback rule: if b(n) = b(n-1), the step size is multiplied by a growth factor P greater than 1. If b(n) is not equal to b(n-1), the step size is multiplied by a decay factor Q less than 1. Common values are P = 1.5 and Q = 0.5, though the optimal choice depends on the signal statistics.

Mathematical Expression

The step size adaptation in ADM is expressed as a recursive update. Let delta(n) denote the step size at sample n. The adaptation rule is as follows. If the current bit b(n) equals the previous bit b(n-1), then delta(n) = P times delta(n-1), where P is the step size multiplier greater than 1. If b(n) is not equal to b(n-1), then delta(n) = Q times delta(n-1), where Q is the step size reducer less than 1. The staircase is then updated as x-hat(n) = x-hat(n-1) plus or minus delta(n) depending on the comparator output.

Bounds are typically placed on the step size to prevent it from growing to infinity or shrinking to zero. These are called delta-min and delta-max. The Signal-to-Quantization-Noise Ratio (SQNR) in ADM is considerably higher than in basic DM for a wide range of input signal frequencies, making ADM significantly more efficient.

Practical Understanding

ADM is used in several real-world voice coding systems. The Continuously Variable Slope Delta (CVSD) modulation, used in military and satellite communication, is a well-known ADM variant. It encodes voice at 16 to 32 kbps with good intelligibility. The adaptive mechanism allows ADM to handle both loud, rapidly changing portions of speech and quiet, nearly constant portions efficiently with the same system.

Compared to PCM, ADM still uses only 1 bit per sample, so the transmitted bit rate remains low. However, because the step size now varies, the receiver must also reconstruct the same step size sequence. This is straightforward since the decoder applies the same adaptation algorithm to the received bit stream to regenerate delta(n) at each step.

Given:
ADM system with P = 1.5, Q = 0.5, initial step size δ(0) = 0.1 V
Transmitted bit stream: 1, 1, 1, 0, 1, 0

Why this formula applies:
Step size adapts based on consecutive or alternating bits.
If b(n) = b(n-1): δ(n) = P × δ(n-1)
If b(n) ≠ b(n-1): δ(n) = Q × δ(n-1)

Formula:
δ(n) = P × δ(n-1)  if b(n) = b(n-1)
δ(n) = Q × δ(n-1)  if b(n) ≠ b(n-1)

Substitution and Calculation:
n=1: b(1)=1, no previous → δ(1) = 0.1 V
n=2: b(2)=1 = b(1)=1 → δ(2) = 1.5 × 0.1 = 0.15 V
n=3: b(3)=1 = b(2)=1 → δ(3) = 1.5 × 0.15 = 0.225 V
n=4: b(4)=0 ≠ b(3)=1 → δ(4) = 0.5 × 0.225 = 0.1125 V
n=5: b(5)=1 ≠ b(4)=0 → δ(5) = 0.5 × 0.1125 = 0.05625 V
n=6: b(6)=0 ≠ b(5)=1 → δ(6) = 0.5 × 0.05625 = 0.028 V

Final Answer: Step sizes are 0.1, 0.15, 0.225, 0.1125, 0.05625, 0.028 V — larger during run of 1s (slope overload region), smaller during alternation (granular noise region).

Exam Tip: In GATE and university exams, ADM is tested on the adaptation rule: consecutive same bits increase step size (slope overload recovery), alternating bits decrease step size (granular noise reduction). Do not confuse P (multiplier, greater than 1) with Q (divider, less than 1).

Mechanism: How ADM Improves Over Basic DM

• Consecutive identical bits (1,1 or 0,0) indicate the staircase is trying to catch a rapidly rising or falling signal, so the step size is increased by factor P to reduce slope overload.
• Alternating bits (1,0,1,0) indicate the signal is nearly flat and the staircase is oscillating, so the step size is reduced by factor Q to reduce granular noise.
• The decoder applies the same P and Q adaptation rules to regenerate delta(n) from the received bit stream before accumulating the staircase.
• ADM outperforms basic DM across a wide range of input signal frequencies and amplitudes, making it suitable for practical voice encoding systems like CVSD.
• Step size bounds delta-min and delta-max prevent numerical instability and ensure the system remains stable for all input levels.

Quick Revision

• ADM solves the fixed step size problem of basic DM by adapting delta(n) at each sample.
• Consecutive same bits trigger step size increase (P greater than 1): slope overload recovery.
• Alternating bits trigger step size decrease (Q less than 1): granular noise reduction.
• Both encoder and decoder use the same adaptation algorithm, so no extra information is needed in the bit stream.
• CVSD is a practical ADM variant used in military voice communications.
• Trap: ADM still transmits 1 bit per sample like basic DM. The adaptation is in step size only, not in the number of bits.
• ADM SQNR is much higher than basic DM SQNR for the same bit rate, especially for wide-band voice signals.