Pulse Code Modulation (PCM) — Communication Systems

Pulse Code Modulation (PCM): The Digital Voice Standard

Pulse Code Modulation is the standard method for digitizing analog signals — particularly voice — and forms the backbone of the entire global telephone network. When you make a phone call, your voice is converted to PCM within milliseconds, transmitted as a stream of bits, and reconstructed back to sound at the other end. Understanding PCM is understanding the foundation of all digital audio.

The Three Steps of PCM

PCM converts an analog signal into digital form through three sequential operations: sampling, quantization, and encoding. Let us walk through each step as if we were building a telephone system from scratch.

Step 1: Sampling

The analog signal is measured at regular intervals. According to the Nyquist theorem, we must sample at least twice the highest frequency present:

fs ≥ 2 × fm(max)

For telephone speech (bandwidth 300-3400 Hz), the standard sampling rate is 8000 samples per second (8 kHz). This is slightly above the Nyquist minimum of 6800 Hz, providing a guard band to prevent aliasing.

Think of it this way: imagine taking rapid snapshots of a bouncing ball. If you take enough snapshots per bounce cycle, you can perfectly reconstruct the ball's trajectory. Too few snapshots, and the motion appears distorted or even reversed (aliasing — like wagon wheels appearing to spin backward in old movies).

Step 2: Quantization

Each sample value must be represented by a finite number of discrete levels. With n bits per sample, we have L = 2ⁿ quantization levels:

• 8 bits → 256 levels (standard telephony)
• 16 bits → 65,536 levels (CD audio)
• 24 bits → 16,777,216 levels (professional audio)

The difference between the actual sample value and the nearest quantization level is called quantization error (or quantization noise). For uniform quantization:

Maximum quantization error = Δ/2 (where Δ = step size = full range / 2ⁿ) Quantization noise power = Δ²/12 SQNR = 6.02n + 1.76 dB (for a full-scale sinusoidal input)

Each additional bit improves SQNR by 6.02 dB. With 8 bits: SQNR = 49.92 dB. With 16 bits: SQNR = 98.08 dB.

Step 3: Encoding

Each quantized sample is converted to its binary representation. The 8-bit code word is transmitted as a serial bit stream at the bit rate:

Rb = fs × n = 8000 × 8 = 64,000 bits/second = 64 kbps

This 64 kbps rate is the fundamental unit of digital telephony — one voice channel. All higher-order multiplexed systems (E1, T1, SONET) are built from multiples of this basic 64 kbps channel.

The Problem with Uniform Quantization

Uniform quantization assigns equal step sizes across the entire amplitude range. This works well for large signals but poorly for small ones. Consider: if the step size is 10 mV, a 1V signal has 0.5% error (negligible), but a 10 mV signal has 50% error (unacceptable).

In speech, quiet sounds are just as important as loud ones — whispering must be intelligible too. Uniform quantization would require 12-13 bits (4096-8192 levels) to achieve acceptable quality across the full dynamic range of speech.

Companding: The Elegant Solution

Companding (compressing + expanding) solves this problem by using non-uniform quantization: smaller step sizes for small amplitudes and larger steps for large amplitudes. This effectively gives quiet sounds more precision while accepting coarser quantization for loud sounds.

Two companding standards exist worldwide:

μ-law (North America, Japan):

F(x) = sgn(x) × ln(1 + μ|x|) / ln(1 + μ) where μ = 255

A-law (Europe, rest of world):

F(x) = sgn(x) × (A|x|)/(1 + ln A) for |x| ≤ 1/A F(x) = sgn(x) × (1 + ln(A|x|))/(1 + ln A) for 1/A < |x| ≤ 1

Where A = 87.6

Both achieve approximately uniform SQNR across a 40 dB dynamic range using only 8 bits — equivalent performance to 12-13 bits of uniform quantization. The improvement is dramatic: companding provides about 24 dB better SQNR for small signals.

PCM Encoding Format

In standard telephony (G.711):

μ-law format (8 bits):

• Bit 1: Sign (1=positive, 0=negative)
• Bits 2-4: Segment code (identifies which of 8 linear segments)
• Bits 5-8: Quantization step within segment (16 steps per segment)

A-law format (8 bits):

• Bit 1: Sign (inverted: 1=positive, 0=negative)
• Bits 2-4: Segment code (8 segments)
• Bits 5-8: Step within segment (16 steps)
• Even bits are inverted for better clock recovery

PCM Bandwidth and Transmission

The minimum bandwidth required to transmit a 64 kbps PCM signal:

Nyquist bandwidth = Rb/2 = 32 kHz (theoretical minimum with ideal sinc pulse shaping)

In practice, with raised-cosine filtering (α = 0.5): Practical bandwidth = 1.5 × 32 = 48 kHz

Time-Division Multiplexing of PCM Channels

Multiple PCM voice channels are combined using Time-Division Multiplexing (TDM):

E1 (European standard):

• 32 time slots × 8 bits × 8000 frames/sec = 2.048 Mbps
• 30 voice channels + 1 signaling + 1 synchronization

T1 (North American standard):

• 24 time slots × 8 bits × 8000 frames/sec = 1.544 Mbps
• 24 voice channels (signaling embedded by "bit robbing")

These are combined into higher hierarchies: E2 (8.448 Mbps), E3 (34.368 Mbps), up to STM-64 (10 Gbps) in modern SONET/SDH networks.

PCM vs. Other Digital Voice Coding

PCM remains the "gold standard" for quality and simplicity. Lower-rate codecs sacrifice some quality for bandwidth efficiency, making them suitable for wireless and VoIP applications where bandwidth is scarce.

Differential PCM (DPCM) and ADPCM

DPCM exploits the correlation between adjacent samples: instead of encoding absolute values, it encodes the difference between predicted and actual samples. Since differences are typically small, fewer bits are needed.

ADPCM (G.726) adapts the step size dynamically based on recent signal characteristics, achieving toll-quality speech at 32 kbps — half the rate of standard PCM.

Practical Example: CD Audio

A CD uses linear PCM with:

• Sampling rate: 44,100 Hz (covers 20 Hz - 20 kHz audio)
• Quantization: 16 bits per sample
• Channels: 2 (stereo)
• Bit rate: 44,100 × 16 × 2 = 1,411,200 bps ≈ 1.41 Mbps
• Dynamic range: 96 dB
• One hour of music: ~635 MB

Key Takeaways

1. PCM digitizes analog signals through sampling (time discretization), quantization (amplitude discretization), and binary encoding.
2. Standard telephone PCM uses 8 kHz sampling, 8-bit quantization with companding, producing 64 kbps per voice channel.
3. Companding (μ-law or A-law) provides uniform quality across the speech dynamic range, equivalent to 12-13 bits of uniform quantization using only 8 bits.
4. The 64 kbps PCM channel is the fundamental building block of the digital telephone network hierarchy.
5. Each additional quantization bit improves SQNR by 6 dB — the fundamental precision-versus-bit-rate trade-off.
6. PCM remains unmatched in simplicity and quality, while compressed codecs (G.729, Opus) sacrifice some quality for dramatically lower bit rates.