Analog Archives

PAM (Pulse Amplitude Modulation) is one of the simplest ways to send digital data. In this post, we start with baseband PAM, look at its spectral efficiency, and then move on to passband PAM — which turns out to be less efficient. To fix that, we explore how QAM (Quadrature Amplitude Modulation) improves efficiency by using both sine and cosine carriers. This leads to I/Q modulation, and the corresponding IQ demodulation with Python code.

The content is based onSec 5.2.1 of [DIG-COMM-BARRY-LEE-MESSERSCHMITT].

Baseband PAM transmission

Consider a simple baseband transmission where the information is sent by modulating a pulse. This can be represented as

$s(t) = \sum_{m=-\infty}^{\infty}a_mg(t-mT)$ , where

$T$ is the symbol period,
$a_m$ is the symbol to transmit,
$g(t)$ is the transmit pulse shaping filter,
$m$ is the symbol index and
$s(t)$ is the output waveform.

Pictorially the same can be represented as,

Figure: Baseband PAM transmission

Minimum bandwidth for ISI free transmission

According to the Nyquist zero ISI criterion, to transmit data without intersymbol interference (ISI), the pulse shaping filter $g(t)$ must satisfy,

$g(nT) = \begin{cases} 1, & n = 0 \\ 0, & \text{otherwise} \end{cases}$

This ensures that each symbol appears cleanly at its own sampling instant without interference from neighboring symbols. For this condition to hold and still remain bandlimited to $\frac{1}{2T}$ , the ideal pulse shape is the sinc function:

$\begin{array}{lll} g(t) &=& \dfrac{\sin\left(\pi t / T\right)}{\pi t / T}, \quad t \in (-\infty, +\infty) \\ &=& \mathrm{sinc}\left(\dfrac{t}{T}\right) \end{array}$

This result comes from the fact that, with ideal sinc pulse shaping, the entire symbol stream can be perfectly reconstructed if the channel passes all frequencies up to $\frac{1}{2T}$ . This is also consistent with the Nyquist–Shannon sampling theorem, which states that to fully reconstruct a signal sampled at $f_s$ , the signal must be bandlimited to $\frac{f_s}{2}$ .

Thus, if we use ideal (bandlimited) pulses like sinc functions, the entire baseband PAM signal can be confined within:

$\text{Bandwidth} = \frac{1}{2T} \text{Hz}$

Figure: Spectrum of baseband PAM (with symbol rate $\frac{1}{T}$ )

Spectral Efficiency

Spectral efficiency is a measure of how efficiently a communication system transmits data over a given bandwidth. It is defined as the number of bits transmitted per second per unit bandwidth, typically expressed in bits/second/Hertz.

$\text{Spectral Efficiency} = \frac{\text{Bit Rate (bits/sec)}}{\text{Bandwidth (Hz)}}$

Assuming that PAM signal $a_m$ uses $N$ distinct amplitude levels, each symbol carries $\log_2(N)$ bits. When each symbol takes $T$ seconds to transmit, the bit rate of the system is:

$\text{Bit Rate} = \frac{\log_2 N}{T} \quad \text{bits/second}$

Substituting the bit rate and bandwidth into the formula for spectral efficiency, we get the spectral efficiency of baseband PAM as.

$\Gamma_{\text{BB-PAM}} = \frac{\log_2 N}{T} \cdot \frac{1}{$\frac{1}{2T}$} = 2 \log_2 N \quad \text{bits/second/Hertz}$

Code

Python code for passband PAM transmission with pulse shaping by rectangular filter and sinc filter.

Passband PAM transmission

Now, consider that the baseband PAM signal is upconverted to a carrier frequency $f_c$ by multiplication by a cosine carrier. The transmitted passband signal is:

$\hat{s}(t)=\sqrt{2}\cos$2\pi f_c t$\sum_{m=-\infty}^{\infty}a_m g\(t-mT)$

The spectrum of the passband PAM is as shown below,

Figure: Spectrum of passband PAM

When the baseband signal is multiplied by $f_c$ , this modulation shifts the entire spectrum up to be centered around the carrier frequency $f_c$ .

The frequency range of the signal becomes:

$f_c - \frac{1}{2T} \) \text{ to } \( f_c + \frac{1}{2T}$

Thus, the total bandwidth needed by passband PAM is,

$\text{Passband Bandwidth} = \left(f_c + \frac{1}{2T}\right) - \left(f_c - \frac{1}{2T}\right) = \frac{1}{T}$

Spectral Efficiency of passband PAM

As earlier, assuming that PAM signal $a_m$ uses $N$ distinct amplitude levels, each symbol carries $\log_2(N)$ bits. When each symbol takes $T$ seconds to transmit, the bit rate of the system is:

$\text{Bit Rate} = \frac{\log_2 N}{T} \quad \text{bits/second}$

Substituting the bit rate and bandwidth into the formula for spectral efficiency, we get the spectral efficiency of passband PAM as,

$\Gamma_{\text{PB-PAM}} = \frac{\log_2 N}{T} \cdot \frac{1}{$\frac{1}{T}$} = \log_2 N \quad \text{bits/second/Hertz}$

Can see that spectral efficiency of passband PAM is half of baseband PAM.

IQ Modulation (QAM for improving spectral efficiency)

As seen earlier, passband PAM suffers from reduced spectral efficiency. To improve spectral efficiency, we exploit the orthogonality of sine and cosine carriers. This leads to Quadrature Amplitude Modulation (QAM), where two independent data streams are modulated:

to a cosine wave (in-phase component, I) $a^I_m$
to a sine wave (quadrature component, Q) $a^Q_m$

Because sine and cosine are orthogonal over a symbol period $\int_T\sin$\omega t$\cdot \cos$\omega t$= 0$ , these two streams do not interfere and can be perfectly separated at the receiver. As a result, QAM doubles the data rate over the same bandwidth compared to passband PAM.

The transmit signal with passband QAM is represented as,

$\bar{s}(t) = \sqrt{2} \cos(2\pi f_c t) \sum_{m=-\infty}^{\infty} a^I_m g(t - mT) - \sqrt{2} \sin(2\pi f_c t) \sum_{m=-\infty}^{\infty} a^Q_m g(t - mT)$

where,

$a^I_m$ and $a^Q_m$ are the in-phase and quadrature data symbols, respectively
$g(t)$ is the transmit pulse shaping filter, and
$f_c$ is the carrier frequency

This signal requires the same bandwidth as passband PAM, however carries twice the information.

Note :

The factor of $\sqrt2$ is included to normalize the total power of the passband QAM signal to unity. As cosine and sine carriers each contribute only half the power $\frac{1}{2}$ when modulating a baseband signal, multiplying each arm by $\sqrt2$ ensures that the modulated in-phase and quadrature components retain their original baseband power.

For notational convenience, we express the passband signal using its complex baseband equivalent,

$\bar{s}(t)=\sqrt{2}\Re \left[ \tilde{s}(t)e^{j2\pi f_c t} \right]$ ,

where

$\tilde{s}(t)=\sum_m \tilde{a}_m g(t - mT)$
$\tilde{a}_m=a^I_m+j\cdot a^Q_m$

This forms the I-Q modulator circuit.

Figure: IQ modulator

IQ demodulation

The received passband signal is split into two branches and multiplied with cosine (in-phase, I) and sine (quadrature, Q) carriers. Each branch is then passed through a matched low-pass filter, typically matched to the transmit pulse shape, to extract the baseband I and Q components. The resulting streams are then independently sampled and demodulated to recover the transmitted symbols.

Downconversion

Let the received passband signal

$\begin{array}{lll}r(t) & = & \bar{s}(t)\\ &=&\sqrt{2} \cos(2\pi f_c t) \sum_{m=-\infty}^{\infty} a^I_m g(t - mT) - \sqrt{2} \sin(2\pi f_c t) \sum_{m=-\infty}^{\infty} a^Q_m g(t - mT)\\ &=&\sqrt{2} \cos(2\pi f_c t) I(t) - \sqrt{2} \sin(2\pi f_c t) Q(t) \end{array}$

where,

$I(t) = \sum_{m=-\infty}^{\infty} a^I_m g(t - mT)$

$Q(t) = \sum_{m=-\infty}^{\infty} a^Q_m g(t - mT)$

To extract the I and Q signals, multiply $r(t)$ with local carrier signals :

In-phase branch:

$\begin{array}{lll} I_{\text{branch}}(t) & = & r(t) \cdot \sqrt{2} \cos(2\pi f_c t) \\ & = & 2I(t)\cos^2(2\pi f_c t) - 2Q(t)\sin(2\pi f_c t)\cos(2\pi f_c t) \\ & = & 2I(t) \left( \frac{1}{2} + \frac{1}{2} \cos(4\pi f_c t) \right) - 2Q(t) \cdot \frac{1}{2} \sin(4\pi f_c t) \\ & = & I(t) + I(t)\cos(4\pi f_c t) - Q(t)\sin(4\pi f_c t) \end{array}$

Similarly, for the quadrature branch:

$\begin{array}{lll} Q_{\text{branch}}(t) & = & r(t) \cdot \sqrt{2} \sin(2\pi f_c t) \\ & = & 2I(t)\cos(2\pi f_c t)\sin(2\pi f_c t) - 2Q(t)\sin^2(2\pi f_c t) \\ & = & 2I(t) \cdot \frac{1}{2} \sin(4\pi f_c t) - 2Q(t) \left( \frac{1}{2} - \frac{1}{2} \cos(4\pi f_c t) \right) \\ & = & I(t)\sin(4\pi f_c t) - Q(t) + Q(t)\cos(4\pi f_c t) \end{array}$

After downconversion, each branch (I and Q) contains the desired baseband signal components, along with unwanted high-frequency terms centered around $2f_c$ .

Low-pass Filtering

To isolate the desired baseband signals, we apply a low-pass filter (LPF) to both the I and Q branches. This filter passes frequencies within the baseband bandwidth (usually $[{-1}/{(2T)}\text{ to } {+1}/{(2T)}]$ where is the symbol period .

Matched Filtering

After low-pass filtering, we apply a matched filter to each of the I and Q branches. The matched filter is designed to maximize the signal-to-noise ratio (SNR) at the symbol sampling instants. It is typically a time-reversed and conjugated version of the transmit pulse shape $g(t)$ .

If the transmit pulse was a rectangular or truncated sinc or raised-cosine shape, the matched filter is also a rectangular or truncated sinc or raised-cosine, respectively. This ensures:

Maximum energy concentration at symbol instants
Minimization or elimination of inter-symbol interference (ISI)

The output of the matched filter is then sampled every seconds (the symbol period) to extract the transmitted symbol values $a^I_m$ and $a^Q_m$ .

This sequence completes the demodulation chain for QAM and ensures that both spectral shaping and ISI suppression are achieved in the receiver.

The IQ demodulator can be visualized as shown in the figure below

Figure: IQ demodulator

Code

Python code for passband qam with rectangular and sinc pulse shaping filter, followed by low pass filtering and matched filtering.

When compared to the transmit symbols, can see that there is a spread in the received symbols. Using raised cosine or root raised cosine filters is a potential approache to mitigate this.

Summary

The post covers the following key aspects

Spectrum of Baseband PAM with a code example
Reduced Spectral Efficiency of Passband PAM
Sending information on orthogonal sine and cosine carriers to improve efficiency (passband QAM)
IQ modulator and corresponding demodulator for Passband QAM with code

Another approach for improving the spectral efficiency of passband PAM is to filter away half the bandwidth (which is not carrying any ‘extra’ information). This is called Single Sideband Modulation (SSB). Anyhow, as I-Q modulation is simpler to implement than circuit for filtering away half the spectrum, I-Q modulation stays. 🙂

Have any questions or feedback? Feel free to drop your feedback in the comments section. 🙂

Reference

[DIG-COMM-BARRY-LEE-MESSERSCHMITT] Digital Communication: Third Edition, by John R. Barry, Edward A. Lee, David G. Messerschmitt

Category: Analog

Image Rejection Ratio (IMRR) with transmit IQ gain/phase imbalance

Modeling phase noise (frequency domain approach)

Transmit spectrum with phase noise

EVM with phase noise

Phase noise power spectral density to Jitter

Continue reading “Phase noise power spectral density to Jitter”

Oscillator phase noise

Thermal noise of RC low pass filter

Noise Figure of cascaded stages

Noise Figure of resistor network

Thermal Noise and AWGN

ADC SNR with clock jitter and quantization noise

IQ imbalance in transmitter

IQ modulation and demodulation

Baseband PAM transmission

Minimum bandwidth for ISI free transmission

Spectral Efficiency

Code

Passband PAM transmission

Spectral Efficiency of passband PAM

IQ Modulation (QAM for improving spectral efficiency)

IQ demodulation

Downconversion

Low-pass Filtering

Matched Filtering

Code

Summary

Reference

2nd order sigma delta modulator