Communication Systems
Comprehensive Study Notes
Based on: Communication Systems Engineering — Proakis & Salehi (2nd Ed.)
Topic Section
Digital Pulse Modulation & Quantization Part I
PCM, TDM & Companding Part II
Delta Modulation & DPCM Part III
Baseband Pulse Transmission & Line Codes Part IV
Noise in Communication Systems Part V
■ PART I: DIGITAL PULSE MODULATION &
QUANTIZATION PROCESS
1.1 Introduction to Analog-to-Digital Conversion
In modern communication systems, analog signals (such as voice, audio, and video) must be
converted into digital form for efficient transmission and storage. The conversion process consists of
three fundamental steps: Sampling, Quantization, and Encoding. Together, these steps form the
foundation of all digital communication systems.
The goal of quantization is to map an infinite set of real numbers (the continuous amplitude of a
sampled signal) into a finite set of discrete levels, while minimizing the distortion introduced.
Because a precise description of an analog source requires an infinite number of bits per sample,
some distortion — called quantization noise — is always present.
1.2 The Quantization Process
In scalar quantization, the real line is partitioned into N disjoint regions (subsets) denoted ℜk, 1 ≤ k ≤
N. To each region ℜk, a single representation point (also called reconstruction level) xk is assigned.
If a sample xi falls in region ℜk, it is replaced by xk.
Since there are N possible quantized levels, log2N bits are sufficient to represent each quantized
value. If N = 2v, then exactly v bits are needed per sample. The price of reducing the bit rate from
infinity (continuous) to v bits per sample is the introduction of distortion — this lost information
cannot be recovered.
Key Note: The quantization function Q(x) = xk for all x in region Rk is nonlinear and non-invertible.
1.3 Midrise and Midtread Quantizers
Two important types of uniform quantizers are defined by the position of their decision boundaries
relative to the origin:
• Midtread Quantizer: Has a quantization region centered at zero. This means zero is itself a
representation point (a quantization output level). The staircase transfer characteristic passes
through the origin. Midtread quantizers have an odd number of output levels, making them
suitable for signals that spend significant time near zero.
• Midrise Quantizer: Has its decision boundary at zero. There is NO output level at exactly zero
— the origin lies at the boundary between two adjacent regions. The staircase rises at the origin.
Midrise quantizers have an even number of output levels (typically N = 2v) and are the most
commonly used in PCM systems.
For a uniform quantizer with N = 2v levels, input range [-Xmax, +Xmax], the step size (quantization
interval) is:
∆ = 2 Xmax / N = 2 Xmax / 2v = Xmax / 2v-1
1.4 Quantization Noise and SQNR
The error introduced by quantization is called quantization noise, defined as the difference between
the original sample X and its quantized version X-hat:
Quantization Error: e = X - Q(X) = X - X-hat
The Signal-to-Quantization-Noise Ratio (SQNR) measures the quality of a quantizer. For a
uniform PCM system with v bits per sample, the SQNR is:
SQNR = 3 × 4v × Pg
where Pg = E[X2] / X2max is the power of the normalized signal. Expressed in decibels:
SQNR|dB ≈ Pg|dB + 6v + 4.8 dB
Key Note: CRITICAL: Each additional bit (increasing v by 1) increases the SQNR by exactly 6 dB.
This is the famous '6 dB per bit' rule, fundamental to all PCM design.
The SQNR in uniform PCM degrades for signals with a large dynamic range because when the
input amplitude is small relative to Xmax, Pg is very small. For instance, speech signals have a
dynamic range of about 40 dB, meaning Pg can vary enormously. This motivates the use of
nonuniform (companded) quantization.
1.5 Optimal Quantizer Design — Lloyd-Max Conditions
The optimal quantizer minimizes the mean-squared distortion D = E[(X - Q(X))2]. The Lloyd-Max
conditions for optimal quantization are:
• Condition 1 (Nearest-neighbor): The quantization region boundaries (decision levels) must be
the midpoints between adjacent representation points: ak = (xk + xk-1) / 2.
• Condition 2 (Centroid): The representation points must be the centroids (conditional means) of
each quantization region: xk = E[X | X ∈ Rk].
Key Note: The Lloyd-Max algorithm iterates between these two conditions to find the optimal
quantizer. For Gaussian sources, Table 6.2 in Proakis & Salehi gives optimal quantization levels for
various values of N.
1.6 Vector Quantization
Vector Quantization (VQ) extends scalar quantization by processing blocks of n source outputs
together, treating them as a single vector in n-dimensional Euclidean space. Instead of quantizing
each sample independently, VQ maps an entire input vector to one of K representative vectors called
codewords stored in a codebook.
The key advantage of VQ is that it can exploit statistical dependencies between samples and
achieve performance approaching the rate-distortion bound as block length n increases. VQ with
n=2 (two-dimensional) already outperforms scalar quantization by using non-rectangular decision
regions better matched to the source distribution. VQ has found widespread application in speech
and image coding.
■ PART II: PCM, COMPANDING (A-LAW & MU-LAW),
AND TIME-DIVISION MULTIPLEXING
2.1 Pulse Code Modulation (PCM)
Pulse-Code Modulation (PCM) is the simplest and most widely used waveform-coding scheme for
converting analog signals to digital form. A PCM system consists of three fundamental stages:
• Sampler: The analog input signal x(t), bandlimited to W Hz, is sampled at the Nyquist rate fs ≥
2W samples/second. A pre-sampling (anti-aliasing) lowpass filter with bandwidth W prevents
aliasing by rejecting all frequency components above W before sampling.
• Quantizer: Each sample is quantized to one of N = 2v discrete levels using either a uniform
quantizer (for uniform PCM) or a nonuniform quantizer (for companded PCM), depending on
the signal characteristics.
• Encoder: Each quantized value is mapped to a binary codeword of length v bits, producing a
serial bit stream for transmission.
For a signal with bandwidth W, sampled at the Nyquist rate fs = 2W and encoded with v bits/sample,
the bit rate of the PCM signal is:
Rb = v × fs = v × 2W = 2vW bits/second
The minimum bandwidth required to transmit this PCM signal (by Nyquist's theorem) is:
BWPCM = v × W (minimum) or BW = v × fs / 2 = vW
Key Note: A PCM system expands the bandwidth of the original signal by a factor of v. For example,
a voice signal of bandwidth W = 4 kHz encoded with v = 8 bits requires a minimum PCM bandwidth of
8 × 4 = 32 kHz. This bandwidth expansion is the price paid for the noise immunity advantages of
digital transmission.
2.2 Companding: Nonuniform Quantization for Improved Dynamic
Range
Companding (a portmanteau of compress and expand) is the standard technique used to overcome
the dynamic range problem in uniform PCM. The key insight is that human speech signals have a
highly non-uniform amplitude distribution — small amplitudes are much more frequent than large
amplitudes.
In a companded PCM system, the process works as follows: At the transmitter, a compressor
(nonlinear element) maps input samples through a nonlinear function g(x), which amplifies small
signals more than large signals. The compressed signal then passes through a uniform quantizer.
At the receiver, the inverse nonlinear function g-1(x), called the expander, restores the original
amplitude distribution.
2.2.1 Mu-Law (µ-Law) Companding — North America and Japan
The µ-law compander is defined by the logarithmic compression characteristic:
g(x) = [log(1 + µ|x|) / log(1 + µ)] × sgn(x), for |x| ≤ 1
where µ controls the amount of compression. A higher µ means more compression of the dynamic
range. The standard used in the United States and Canada (DS-1 telephone standard) employs µ =
255 with 128 quantization levels (v = 7 bits). The use of the µ-law compander improves system
performance by approximately 24 dB compared to uniform PCM for speech signals with large
dynamic range.
2.2.2 A-Law Companding — Europe and International Systems
The A-law compander (used in Europe and most international standards) is defined by a piecewise
function:
g(x) = A|x| / (1 + log A) × sgn(x), for 0 ≤ |x| ≤ 1/A g(x) = [1 + log(A|x|)] / (1 + log A) × sgn(x),
for 1/A ≤ |x| ≤ 1
The standard A-law parameter is A = 87.56. The A-law and µ-law companders provide comparable
performance for speech coding. The A-law has a slightly simpler linear region near the origin, which
makes it easier to implement and less susceptible to noise at very low amplitudes. Both laws provide
a nearly constant SQNR over a wide dynamic range of input signal amplitudes.
Key Note: Key comparison: µ-law is used in North America (PCM standard DS-1, T1 carrier), while
A-law is used in Europe (CEPT standard, E1 carrier). Both use 8 bits/sample and 8000
samples/second, giving a bit rate of 64 kbits/second per voice channel.
2.3 Time-Division Multiplexing (TDM)
Time-Division Multiplexing (TDM) is a method of combining multiple digital signals into a single
high-speed data stream for transmission over a common channel. TDM is the natural multiplexing
scheme for PCM systems.
In TDM, a given time interval Tf is defined as a frame. Each frame is subdivided into N equal time
slots, each of duration Tf/N. Each subscriber (user) is assigned one time slot per frame. In PCM
telephony, each user transmits one 8-bit sample per frame.
The North American TDM Hierarchy (based on the DS standard) is as follows:
Level Designation Number of Voice Channels Bit Rate
1 DS-1 (T1) 24 1.544 Mbits/sec
2 DS-2 (T2) 96 (4 × DS-1) 6.312 Mbits/sec
3 DS-3 (T3) 672 (7 × DS-2) 44.736 Mbits/sec
The DS-1 (T1) signal carries 24 voice channels. Each channel uses 8 bits/sample at 8000
samples/sec = 64 kbits/sec. The T1 bit rate of 1.544 Mbits/sec = 24 × 64 kbits/sec + 8 kbits/sec
(framing bits). TDM is fundamentally different from Frequency-Division Multiplexing (FDM), which
stacks multiple signals in frequency rather than time.
■ PART III: DELTA MODULATION, ADAPTIVE DM,
LINEAR PREDICTION & DPCM
3.1 Differential PCM (DPCM)
In standard PCM, each sample is quantized and encoded independently. However, for signals like
speech, adjacent samples are highly correlated. Differential PCM (DPCM) exploits this correlation by
transmitting only the difference between the current sample and a predicted value, rather than the
sample itself.
The key advantage: the difference (prediction error) has a much smaller dynamic range than the
original signal. Therefore, fewer bits are needed to quantize it to the same fidelity, achieving a lower
bit rate than standard PCM. The predictor typically uses the previous quantized output as the
prediction:
Predicted value: X-hatn = Yn-1 (first-order predictor)
Prediction error: en = Xn - X-hatn
Quantized error: Yn = en + quantization error
Key Note: A crucial property of DPCM: the quantization error between Xn and its reproduction equals
the quantization error between the prediction error and its quantized version. DPCM can achieve
better performance than PCM at lower bit rates, particularly for speech signals.
3.2 Linear Prediction
Linear Prediction uses a more sophisticated multi-step predictor based on p previous quantized
samples. The predicted value is a weighted linear combination:
X-hatn = a1Yn-1 + a2Yn-2 + ... + apYn-p
where {ak} are the predictor coefficients chosen to minimize the mean-squared prediction error.
The optimal predictor coefficients satisfy the Wiener-Hopf equations (Yule-Walker equations)
involving the autocorrelation of the source signal.
Linear Predictive Coding (LPC) is a powerful analysis-synthesis technique widely used in speech
coding. An LPC vocoder uses a model of the vocal tract as an all-pole filter and estimates its
coefficients every frame (typically every 10–30 ms). By transmitting only the predictor coefficients and
the excitation signal (rather than the waveform), LPC achieves bit rates of about 2400–9600 bits/sec
for speech — far below the 64,000 bits/sec of standard PCM. LPC with vector quantization has been
adopted as the standard for speech compression in mobile (cellular) telephone systems.
3.3 Delta Modulation (DM)
Delta Modulation (∆M) is the simplest possible form of DPCM, using a 1-bit (two-level) quantizer with
levels ±∆. It can be viewed as DPCM with a first-order predictor and a single-bit quantizer. The output
of the ∆M encoder is a simple binary sequence that tracks whether the current sample is above or
below the previous reconstruction.
Because only 1 bit per sample is transmitted, the quantization error is very high unless the signal
changes very slowly. To maintain accuracy, ∆M samples at a rate much higher than the Nyquist
rate. Even though the individual sample resolution is coarse (1 bit), the high sampling rate enables
good reconstruction.
The reconstruction is performed by a simple integrator (accumulator):
X-hatn = Σi=0n Yi ≈ ∆ × (number of +1's minus number of -1's up to time n)
3.3.1 Two Types of Delta Modulation Distortion
• Slope Overload Distortion (large ∆): Occurs when the input signal changes too rapidly for the
step-size ∆ to track it. The quantizer output lags behind the input, producing a triangular
reconstruction wave that cannot follow steep slopes. The condition to avoid slope overload is:
|dx(t)/dt| ≤ ∆ × fs.
• Granular Noise (small ∆): Occurs when the input signal is nearly constant. The reconstructed
waveform oscillates around the signal value in a staircase fashion, producing a buzzing/granular
noise. The reconstruction hunts around the true value with amplitude ±∆.
Key Note: There is a fundamental trade-off: large ∆ reduces slope overload but increases granular
noise, and vice versa. The optimal ∆ depends on the signal characteristics. This motivates Adaptive
Delta Modulation.
3.4 Adaptive Delta Modulation (ADM)
Adaptive Delta Modulation solves the fixed step-size problem by dynamically adjusting the step
size ∆ based on the recent behavior of the input signal:
• When the input signal is changing rapidly (slope overload condition detected — consecutive
output bits have the same sign), the step size is increased to track the rapid changes.
• When the input signal is slowly varying (granular noise condition — output bits alternate in
sign), the step size is decreased to reduce granular noise.
ADM achieves a better signal-to-noise ratio than fixed ∆M for a wide variety of input signals. The step
size adaptation can be implemented very simply — just by examining the last two output bits. ADM is
used in some satellite communication systems and military communication links.
3.5 Vocoders and Video Compression
Vocoders (Voice Coders) are analysis-synthesis systems that model the speech production process
rather than encoding the speech waveform directly. The key types are:
• LPC Vocoder: Models the vocal tract as an all-pole (AR) filter driven by either a periodic pulse
train (voiced sounds) or random noise (unvoiced sounds). Transmits model parameters: predictor
coefficients, pitch period, voiced/unvoiced flag, and gain. Achieves bit rates of 2400 bits/sec. Used
in military secure voice systems.
• CELP (Code Excited Linear Prediction): Improves LPC by using a codebook of excitation
sequences. The encoder searches through a codebook to find the excitation that best matches
the original speech. Achieves toll-quality speech at 4800–9600 bits/sec.
Video Compression (JPEG, MPEG) uses the Discrete Cosine Transform (DCT) as its core tool. In
JPEG, a picture is divided into 8×8 pixel blocks. The DCT of each block concentrates the signal
energy into a few low-frequency coefficients, which are then quantized (the JPEG quantization table
assigns coarser quantization to higher-frequency coefficients) and encoded using run-length coding
(for the many near-zero high-frequency coefficients) and Huffman coding. MPEG extends this by also
exploiting temporal redundancy between successive video frames.
■ PART IV: BASEBAND PULSE TRANSMISSION
4.1 Line Codes and Power Spectral Density
A line code is the mapping of binary data {0, 1} into a specific electrical waveform for transmission
over a baseband channel. The choice of line code affects bandwidth efficiency, synchronization
capability, error detection ability, and DC component. The most important line codes are:
Line Code Description DC Content Self-Clocking
Unipolar NRZ 1 → +A, 0 → 0 Yes (A/2) Poor
Polar NRZ 1 → +A, 0 → -A None (if equal prob.) Poor
NRZ-I (NRZI) Transition for 1, none for 0 None Moderate
Bipolar / AMI 1 alternates ±A, 0 → 0 None always Good
Manchester 1 → +A to -A, 0 → -A to +A None Excellent
Miller Code Complex transition rules None Good
4.1.1 NRZ and NRZI Signaling
NRZ (Non-Return-to-Zero): Identical to binary PAM where bit 1 is represented by amplitude +A and
bit 0 by amplitude −A. The signal level remains constant throughout the bit period T (no return to
zero). NRZ has a lowpass power spectrum with a null at f = 1/T and contains a DC component.
NRZI (Non-Return-to-Zero-Inverse): A transition from one level to another occurs only when the
data bit is 1; no transition occurs for bit 0. NRZI is state-dependent and can be viewed as NRZ
preceded by a precoding operation. NRZI is used in magnetic recording and CD systems.
4.1.2 Bipolar (AMI) Signaling
In Alternate Mark Inversion (AMI) or bipolar signaling, binary 0 is represented by zero amplitude,
while binary 1 alternates between +A and −A. The key properties of AMI are:
• No DC component: Due to the sign alternation of 1's, the average value is zero regardless of
the data pattern. The PSD of AMI has a zero at f = 0.
• Error detection: A violation of the alternation rule (two consecutive 1's with the same polarity)
immediately signals a transmission error.
• Long string of zeros problem: A long sequence of 0's produces no transitions, making timing
recovery difficult. This is addressed by B3ZS (Bipolar with 3-Zero Substitution) or B8ZS codes.
Key Note: PSD (Power Spectral Density) of Bipolar/AMI: S(f) = 2A2T sin2(πfT). This has nulls at f = 0
(no DC) and f = 1/T, and a peak near f = 1/(2T).
4.2 Intersymbol Interference (ISI)
Intersymbol Interference (ISI) is the contamination of the current symbol by energy from
neighboring symbols, caused by the finite bandwidth of the communication channel. A bandlimited
channel acts as a filter — it spreads the transmitted pulses in time, causing them to overlap at the
receiver.
In a general baseband digital communication system, the received signal at the output of the matched
filter, sampled at time t = mT, is:
ym = x(0)am + Σn≠m x((m-n)T) an + noise
The second term — the sum over n ≠ m — is the ISI. It represents the contribution of all other
transmitted symbols {an, n ≠ m} to the current sample. ISI causes degradation in the probability of
error of the system.
ISI is caused by: (1) amplitude distortion in the channel (different frequencies attenuated unequally),
(2) phase (delay) distortion (different frequencies experience different group delays), and (3)
insufficient channel bandwidth. The eye pattern is the standard tool for visualizing ISI — it is obtained
by superimposing successive segments of the waveform on an oscilloscope. A wide-open eye means
low ISI.
4.3 Nyquist Criterion for Zero ISI
The most important result in baseband digital transmission design is the Nyquist pulse-shaping
criterion: it gives the necessary and sufficient conditions on the overall system transfer function X(f)
= GT(f) × C(f) × GR(f) such that the sampled output has zero ISI.
Nyquist's Theorem: A necessary and sufficient condition for x(t) to satisfy x(nT) = 1 (n=0) and x(nT)
= 0 (n ≠ 0) — zero ISI condition — is that:
Σm X(f + m/T) = T, for all f in [-1/(2T), 1/(2T)]
This means the sum of frequency-shifted versions of X(f) must be constant. Physical interpretation:
The spectrum X(f) must be designed so that its aliases (shifted by multiples of 1/T) sum to a constant.
This is achieved when X(f) has a special symmetry about f = 1/(2T).
Ideal Nyquist Bandwidth: The absolute minimum bandwidth for ISI-free transmission at symbol rate
1/T is W = 1/(2T), corresponding to the ideal rectangular spectrum. However, this pulse — the sinc
function — has infinitely long tails that decay only as 1/t. A small timing error causes a
non-converging ISI series, making the ideal sinc pulse impractical.
4.3.1 Raised Cosine Spectrum — Practical Zero-ISI Design
The raised cosine spectrum is the most widely used practical pulse shape satisfying Nyquist's
criterion. It is defined by:
Xrc(f) = T, for |f| ≤ (1-α)/(2T) Xrc(f) = T/2 × [1 + cos(πT/α(|f| - (1-α)/(2T)))], for (1-α)/(2T) ≤ |f| ≤
(1+α)/(2T)
where α is the rolloff factor (0 ≤ α ≤ 1). Key properties:
• α = 0: Ideal rectangular spectrum (minimum bandwidth W = 1/(2T)), sinc(t/T) pulse —
impractical due to slow 1/t decay.
• α = 1: Maximum bandwidth W = 1/T (twice the minimum), but pulses decay as 1/t3 — much
more robust to timing errors. Occupied bandwidth = 2/T.
• 0 < α < 1: Practical compromise. The bandwidth is (1+α)/(2T). The pulse decays fast enough for
timing jitter tolerance.
Key Note: In practice, the raised cosine filtering is split between the transmitter filter (square root
raised cosine: SRRC) and receiver filter (SRRC). The cascade of two SRRC filters gives the overall
raised cosine spectrum, achieving the matched filter condition simultaneously with zero ISI.
4.4 Controlled ISI (Partial Response Signaling)
Partial response signaling (also known as controlled ISI or correlative coding) is a technique that
deliberately introduces a controlled, known amount of ISI between adjacent symbols, enabling
signaling at the Nyquist rate within a bandwidth of W = 1/(2T) while avoiding the impractical ideal sinc
pulse.
The most common partial response scheme is duobinary (Class I): Each transmitted symbol is the
sum of the current and previous symbol: yn = an + an-1. If an ∈ {±1}, then yn ∈ {-2, 0, +2} — a 3-level
output from 2-level input. The receiver uses precoding to avoid error propagation in detection.
4.5 Differential Encoding
Differential encoding maps information bits into transitions rather than absolute levels. The encoder
computes: dn = an XOR dn-1. The key advantage: the receiver does not need to know the absolute
phase reference — it only needs to detect whether a transition occurred. This is particularly useful in
systems where phase ambiguity (180° phase uncertainty) exists, such as PSK systems.
4.6 Baseband Data Transmission in White Gaussian Noise —
Probability of Error
For a binary baseband PAM system transmitting symbols ±A through an AWGN channel with
two-sided noise PSD N0/2, the optimum receiver is a matched filter (or correlator) followed by a
threshold comparator. The probability of error is:
Pe = Q(√(2Eb/N0))
where Eb is the energy per bit and Q(·) is the Q-function (tail probability of the standard Gaussian
distribution). For M-ary PAM with M equally spaced signal levels, the probability of symbol error is:
Ps = 2(M-1)/M × Q(√(6 log2M × Eb / ((M2-1) N0)))
Key Note: The matched filter maximizes the output SNR at the sampling instant. For binary antipodal
signaling, the minimum probability of error achievable by any receiver is Pe = Q(√(2Eb/N0)), and the
matched filter achieves this bound.
4.7 Band-Limited Nature of Channels and M-ary Transmission
Real channels have finite bandwidth W. To transmit data at a bit rate Rb bits/second through a
channel of bandwidth W, we can use M-ary signaling with M = 2k levels, where each symbol carries
k = log2M bits. The symbol rate (baud rate) is then:
fs = Rb / log2M = Rb / k [symbols/second]
By Nyquist's theorem, a channel of bandwidth W can support a symbol rate of at most 2W
symbols/second (with ideal Nyquist filtering). Therefore, the maximum achievable bit rate is Rb,max =
2W log2M bits/second. Increasing M allows higher bit rates within the same bandwidth, but at the cost
of increased error probability for the same signal power (because signal points are closer together).
■ PART V: NOISE IN COMMUNICATION SYSTEMS
5.1 Receiver Model and Figure of Merit
When evaluating the performance of a communication receiver in the presence of additive white
Gaussian noise (AWGN), the figure of merit is the ratio of output signal power to output noise power
— the output SNR. To fairly compare different modulation systems, we normalize all comparisons to
a baseband reference.
The standard receiver model assumes: (1) the received signal plus noise enters a noise-limiting
(bandpass) filter that passes the signal bandwidth but rejects out-of-band noise, (2) the output is
passed to the demodulator appropriate for the modulation scheme, (3) the demodulated signal
passes through a lowpass post-detection filter of bandwidth W. The noise model throughout is
AWGN with two-sided PSD N0/2.
The figure of merit for a modulation system is defined as:
Figure of Merit = (S/N)output / (S/N)baseband = (output SNR) / (SNR for equivalent baseband
system)
For the baseband reference system, the received signal passes directly through a lowpass filter of
bandwidth W. The output SNR is:
(S/N)baseband = PR / (N0W)
where PR is the received signal power. This baseline SNR is used as the denominator in comparing
all modulation systems.
5.2 Noise in DSB-SC AM Systems
In Double-Sideband Suppressed-Carrier AM (DSB-SC), the transmitted signal is u(t) =
Acm(t)cos(2πfct). At the receiver, coherent demodulation multiplies by cos(2πfct) and lowpass filters:
After demodulation, the output signal is (1/2)Acm(t) and the output noise is (1/2)nc(t), where nc(t) is
the in-phase noise component. The output SNR is:
(S/N)DSB-SC = Ac2Pm / (2N0W) = PR / (N0W) = (S/N)baseband
Key Note: DSB-SC provides exactly the same SNR as a baseband system — no SNR advantage
over baseband. The figure of merit for DSB-SC = 1. The bandwidth efficiency is poor (twice the
message bandwidth), but DSB-SC is the benchmark for AM comparisons.
5.3 Noise in SSB AM Systems
For Single-Sideband AM (SSB), only one sideband (upper or lower) is transmitted:
u(t) = Acm(t)cos(2πfct) ± Acm-hat(t)sin(2πfct)
With coherent demodulation, the output SNR for SSB is:
(S/N)SSB = Ac2Pm / (2N0W) = (S/N)baseband
SSB also has a figure of merit = 1, the same as DSB-SC. However, SSB has a major bandwidth
advantage: it occupies only W Hz bandwidth (half that of DSB), making SSB bandwidth-efficient.
SSB is used in high-frequency (HF) radio communications and frequency-division telephone systems.
5.4 Noise in Conventional (Full-Carrier) AM Systems
In conventional AM, u(t) = Ac[1 + amn(t)]cos(2πfct), where a is the modulation index and mn(t) is the
normalized message.
For synchronous (coherent) demodulation of conventional AM, the output SNR is:
(S/N)AM,sync = (a2Pmn / (1 + a2Pmn)) × (S/N)baseband
Since the factor a2Pmn / (1 + a2Pmn) < 1 always, conventional AM always has a lower SNR than a
baseband system with the same received power. The fraction of received power wasted in the
carrier is 1/(1 + a2Pmn). For typical speech (Pmn ≈ 0.1) with modulation index a = 0.85, the SNR loss
compared to baseband is about 13 dB.
5.5 Threshold Effect in AM Envelope Detection
For envelope (non-coherent) detection of conventional AM, when the SNR is high (above
threshold), the envelope detector output is approximately:
Vr(t) ≈ Ac[1 + amn(t)] + nc(t)
giving approximately the same SNR as coherent detection. However, when the SNR drops below a
threshold (noise envelope becomes comparable to signal envelope), the envelope detector output is
dominated by noise and the message signal is effectively lost — this is called the threshold effect.
Below threshold, performance degrades catastrophically, far worse than coherent detection.
5.6 Noise in Frequency Modulation (FM) Systems
FM provides a fundamental trade-off: bandwidth for SNR improvement. The FM discriminator
output contains noise with a parabolic (quadratic) power spectrum — higher frequencies in the
message band experience more noise. The output SNR for FM is:
(S/N)FM = 3 β2(β + 1) Pmn × (S/N)baseband
where β = ∆f / W is the modulation index (∆f = peak frequency deviation, W = message bandwidth).
The FM improvement factor over baseband is 3β2(β + 1)Pmn. For large β, this grows as 3β3Pmn,
meaning FM SNR improves as the cube of the bandwidth expansion.
Key Note: Example: For β = 5 (wideband FM, Pmn = 1/3), the FM SNR improvement over baseband is
3 × 25 × 6 × 1/3 = 150 = 21.8 dB! Commercial FM radio uses β = 5, giving dramatic noise reduction
compared to AM broadcasting.
5.6.1 FM Threshold Effect and Threshold Reduction
Just like AM envelope detection, FM also exhibits a threshold effect. The FM demodulator
(discriminator) operates correctly only when the carrier SNR (received SNR before demodulation) is
above threshold. Below threshold, the noise periodically makes the phase of the received signal slip
by 2π, causing random spike impulsive noise at the discriminator output — the signal is lost in noise.
The FM threshold condition is approximately:
(S/N)threshold = (S/N)baseband,threshold ≈ 20(β + 1)
FM Threshold Reduction techniques allow operation at lower received SNR values. The two main
approaches are:
• FMFB (FM with Feedback): Uses a feedback loop (PLL) to reduce the instantaneous frequency
deviation seen by the discriminator, effectively allowing the use of a narrowband discriminator
even for wideband FM. This reduces the threshold by approximately 5–7 dB.
• PLL Demodulator: A phase-locked loop tracks the instantaneous phase of the FM signal. The
PLL has a lower threshold than the conventional discriminator because it is a phase-coherent
device. The threshold reduction of a PLL-based FM demodulator is typically 3–5 dB.
5.6.2 Pre-emphasis and De-emphasis in FM
Since FM noise has a parabolic power spectrum (noise increases with frequency as f2), the
high-frequency components of the message suffer disproportionately. Pre-emphasis at the
transmitter boosts the high-frequency components of the message signal before modulation.
De-emphasis at the receiver applies the inverse filter to restore the original signal, simultaneously
attenuating the high-frequency noise. Commercial FM broadcasting uses a pre-emphasis time
constant of 75 µs in North America (50 µs in Europe), providing about 13 dB of SNR improvement for
speech signals.
5.7 Noise in Phase Modulation (PM) Systems
In Phase Modulation (PM), the instantaneous phase is proportional to the message: φ(t) = kpm(t).
The output SNR for PM is:
(S/N)PM = kp2 Pm × (S/N)baseband / W2
Unlike FM, PM noise has a flat power spectrum (constant with frequency within the message
bandwidth). For sinusoidal modulation, βPM = kp max|m(t)| and the PM SNR is similar to FM.
However, for speech and music signals, FM is generally superior to PM because FM naturally
provides more bandwidth to low-frequency signal components.
5.8 Noise in PCM Systems — Quantization Noise plus Transmission
Noise
In a PCM system transmitted over a noisy channel, two sources of distortion are present:
• Quantization distortion Dq: Due to the limited number of quantization levels v. Dq = X2max / (3
× 4v).
• Transmission distortion Dt: Due to bit errors caused by channel noise. When a bit error occurs
in the binary codeword, it introduces an additional error in the reconstructed sample that depends
on which bit position is in error (LSB errors cause small distortion, MSB errors cause large
distortion).
Assuming a bit error probability Pb and that each v-bit PCM codeword has at most one bit error, the
total distortion is:
Dtotal ≈ (X2max / 3) × (4-v + 4Pb)
The resulting SNR at the PCM receiver output is:
SNRPCM = 3 X2 / (X2max (4-v + 4Pb))
Key observations: (1) For small Pb, the SNR grows exponentially with v (with bandwidth expansion
factor Wc/Ws = v). (2) Once Pb exceeds about 4-v/4, transmission errors dominate and increasing v
provides no further benefit. (3) PCM trades bandwidth for SNR exponentially, which is more efficient
than FM (which trades bandwidth for SNR as a cubic function).
Key Note: Comparison: FM SNR grows as β3 (cubic in bandwidth expansion). PCM SNR grows as 4v
= 22v (exponential in bandwidth expansion factor v = Wc/Ws). PCM is far more bandwidth efficient
than FM for high-quality (high SNR) applications.
These study notes are based on Communication Systems Engineering by Proakis & Salehi (2nd Ed.). All key formulas
and concepts have been derived from the textbook and are presented for academic study purposes.