June 18, 2020 On a cold morning in August 2015, I narrowly missed a train to my office in Melbourne city. With nothing else to do in the next 20 minutes, my mind wandered towards an intuitive view of complex numbers, something that has puzzled me since long. In particular, I wanted to seek answers to the following questions. (a) What is the role of the number $\sqrt{-1}$ in mathematics? What sets it apart from other impossible numbers, e.g., a number $k$ such that $|k|=-1$? (The origins of this question might lie in how I cut apple slices for my

Continue reading# Category: DSP

Digital Signal Processing

## Proof of Poisson Sum Formula

The Poisson sum formula was discovered by the French mathematician and physicist Siméon Denis Poisson. It has several applications in digital signal processing, among which our concern is the periodic summation of modulated pulses in digital communication systems. Assume that $p(t)$ is a pulse shape (or any continuous-time function if you are not familiar with digital communications) and $P(f)$ is its Fourier Transform. The pulse is sampled at a rate of $f_s$ to produce its discrete version $p(nT_s)$ where $T_s=1/f_s$ is the duration between two samples. The Poisson summation formula relates these two quantities as \begin{equation}\label{equation-poisson-sum-formula} \frac{1}{T_s}\sum _{k=-\infty}^{\infty} P\left(f+\frac{k}{T_s}\right) =

Continue reading## Antenna Arrays: Whole is More than its Parts

In a previous article, we described how multiples antennas can be viewed as sampling the signal in space domain in a direct analogy to sampling the signal in time domain. Filters in discrete-time domain are usually defined by numerical values in the time axis. Space allows three-dimensional freedom. A filter in discrete space can have its elements placed anywhere in the space. This sounds very complicated! But we will show it is not. A Simple Time Domain Filter: 2-Tap FIR The simplest discrete FIR filter has only two samples of unit magnitude. Assuming unit magnitude and zero phase for both

Continue reading## Discrete Fourier Transform (DFT) as a Filter Bank

We have discussed before what a Discrete Fourier Transform (DFT) is and how to find the DFT of some commonly used signals. Here, we will see how a DFT acts as a (crude) bank of filters that can pass the signal contents around a desired frequency while blocking the rest. Let us start with the definition of the DFT. \begin{equation*} \begin{aligned} S_I[k]\: &= \sum \limits _{n=0} ^{N-1}\left[ s_I[n] \cos 2\pi\frac{k}{N}n + s_Q[n] \sin 2\pi\frac{k}{N}n\right] \\ S_Q[k] &= \sum \limits _{n=0} ^{N-1}\left[ s_Q[n] \cos 2\pi\frac{ k}{N}n – s_I[n] \sin 2\pi\frac{k}{N}n\right] \end{aligned} \end{equation*} for each $k$. In complex notation, this DFT is

Continue reading## 3 Ways to Approximate atan2( ) in Hardware

When we think about signal processing, the focus is usually magnitude response of a system. However, in several DSP applications, the signal phase holds as much, if not more, significance as the magnitude response. For example, in digital FM demodulation, carrier phase synchronization and RF ranging, the phase (found through arctangent in four quadrants) of a complex signal needs to be computed by an FPGA or a DSP for further processing. In image processing applications, such an operation is also required to calculate the gradient orientations used in several popular feature descriptors like the Scale-Independent Feature Transform (SIFT) or the

Continue reading