Delta Modulation

Delta Modulation: Simplifying Digital Voice Encoding

Delta Modulation (DM) is a beautifully simple approach to converting analog signals into digital form. While Pulse Code Modulation (PCM) quantizes the absolute amplitude of a signal at each sample, delta modulation takes a different philosophy — it only encodes the *change* (delta) between consecutive samples. This single insight leads to a system that needs only one bit per sample, making it one of the simplest analog-to-digital conversion techniques ever devised.

The Core Idea: Predicting and Correcting

Think of it this way: imagine you are describing someone's position as they walk along a hilly path. PCM would be like calling out their exact altitude at every step: "152 meters, 153 meters, 155 meters, 154 meters..." Delta modulation would be like saying: "up, up, up, down..." — just one bit indicating the direction of change.

The delta modulator maintains a "staircase approximation" of the input signal. At each sampling instant, it compares the actual input with the current staircase level and makes a simple binary decision:

• If the input is higher than the current approximation → output "1" (step up)
• If the input is lower than the current approximation → output "0" (step down)

Each step has a fixed size Δ (delta), hence the name.

Mathematical Description

The system operates at sampling frequency fs (much higher than Nyquist rate) and produces one bit per sample:

Output bit: e(nTs) = sign[x(nTs) - x̂(nTs)]

Staircase update: x̂((n+1)Ts) = x̂(nTs) + Δ × e(nTs)

Where:

• x(nTs) = input signal at sample n
• x̂(nTs) = staircase approximation at sample n
• Δ = step size
• e(nTs) = +1 or -1 (encoded as 1 or 0)

The bit rate equals the sampling frequency: Rb = fs (one bit per sample).

Block Diagram

Delta Modulator (Transmitter):

1. Comparator — Compares input x(t) with feedback signal x̂(t)
2. Sampler — Samples the comparator output at frequency fs
3. Quantizer — One-bit quantizer (just the sign)
4. Accumulator (integrator) — Adds ±Δ to previous approximation

Delta Demodulator (Receiver):

1. Accumulator — Identical to transmitter's accumulator, reconstructs staircase
2. Low-pass filter — Smooths the staircase into a continuous waveform

The beauty of this system is that the demodulator is trivially simple — just an integrator and a filter. No complex D/A converter is needed.

The Two Enemies: Slope Overload and Granular Noise

Delta modulation faces two fundamental error sources that pull the designer in opposite directions:

Slope Overload occurs when the input signal changes faster than the staircase can follow. If the maximum slope of the input exceeds the staircase's maximum tracking rate:

Maximum trackable slope = Δ × fs

For a sinusoidal input x(t) = A × sin(2πfm × t), the maximum slope is 2πfm × A. To avoid slope overload:

Δ × fs ≥ 2πfm × A

Δ(min) = 2πfm × A / fs

When slope overload occurs, the staircase falls far behind the input, creating large distortion that sounds like "clipping" in audio.

Granular Noise occurs during relatively flat portions of the signal. Even when the input is constant, the staircase hunts above and below the true value by ±Δ, producing a triangular "chatter" noise. This noise power is:

Granular noise power ≈ Δ²/3

The dilemma is clear: large Δ prevents slope overload but increases granular noise; small Δ reduces granular noise but causes slope overload on fast-changing signals.

Choosing the Step Size

The optimal step size balances both error types. For a voice signal with bandwidth fm and sampling frequency fs:

Δ(optimal) ∝ 2πfm × A / fs

For good speech quality, DM typically uses:

• Sampling rate: 4-8 times the Nyquist rate (32-64 kHz for 4 kHz voice)
• Step size: chosen to minimize total distortion

The Signal-to-Quantization-Noise Ratio (SQNR) for an optimally-designed DM:

SQNR ≈ (3fs³) / (8π²fm³)

In dB: SQNR (dB) ≈ 10log₁₀(3fs³ / 8π²fm³)

Increasing sampling frequency by a factor of 2 improves SQNR by 9 dB (compared to 6 dB per bit in PCM).

Adaptive Delta Modulation (ADM)

Adaptive Delta Modulation solves the slope-overload/granular-noise trade-off by varying the step size dynamically:

• During rapidly-changing signal segments → increase Δ (prevent slope overload)
• During slowly-changing segments → decrease Δ (reduce granular noise)

Detection logic: If several consecutive bits are the same (e.g., 1,1,1,1), the staircase is falling behind → increase Δ. If bits alternate (1,0,1,0), the staircase is hunting around the signal → decrease Δ.

Common adaptation algorithms:

• Continuously Variable Slope DM (CVSD): Step size controlled by a leaky integrator counting run lengths
• Digitally-controlled ADM: Step multiplied by constant (e.g., 1.5×) for same-bit runs, divided for alternating bits

CVSD is used in military communication (MIL-STD-188-113) and Bluetooth voice transmission because of its robustness to bit errors and channel conditions.

DM vs. PCM: A Comparison

Key advantage of DM: A single bit error causes only a small step error (±Δ), while in PCM a single bit error in the MSB causes a massive amplitude error. This makes DM inherently more robust in noisy channels.

Sigma-Delta Modulation: DM's Modern Descendant

The sigma-delta (ΣΔ) modulator is a sophisticated evolution of basic DM used in modern audio ADCs (your smartphone, laptop, and headphones all use sigma-delta converters). It integrates the input signal before delta modulation, which shapes the quantization noise spectrum — pushing noise energy to higher frequencies where it can be filtered out.

Sigma-delta ADCs achieve 16-24 bit resolution from a 1-bit quantizer by:

• Oversampling at 64-256× Nyquist rate
• Noise shaping to move quantization noise out of the signal band
• Digital decimation filtering to produce the final high-resolution output

Practical Example

Design a DM system for telephone-quality speech (300 Hz - 3400 Hz):

• Maximum frequency: fm = 3400 Hz
• Choose sampling rate: fs = 8 × 2 × 3400 = 54,400 Hz ≈ 56 kHz
• Maximum signal amplitude: A = 1V peak
• Minimum step size (slope overload): Δ ≥ 2π × 3400 × 1 / 56000 ≈ 0.38V
• Granular noise power: σ² ≈ 0.38²/3 ≈ 0.048 V²
• SQNR ≈ 10log₁₀(0.5/0.048) ≈ 10 dB (basic DM — poor!)

This shows why adaptive DM or higher sampling rates are essential for acceptable quality.

Key Takeaways

1. Delta modulation encodes only the change between consecutive samples, producing exactly one bit per sample.
2. The two fundamental distortion sources — slope overload and granular noise — impose conflicting requirements on step size.
3. Adaptive DM (CVSD) dynamically adjusts step size, dramatically improving quality over fixed-step DM.
4. DM is inherently robust to channel bit errors since each error only shifts the staircase by one step.
5. SQNR improves by 9 dB for each doubling of sampling rate — requiring higher oversampling than PCM.
6. Sigma-delta modulation extends DM principles to achieve high-resolution audio conversion in modern electronics.