Wave Walker DSP

DSP Algorithms for RF Systems

New Book!

Now for sale: The third edition of DSP for Beginners: Simple Explanations for Complex Numbers! Includes a new chapter on sampling.

Phase relationship between sine and cosine

Designing an FIR High Pass Filter with Remez

September 23, 2021

Introduction

While low-pass filtering (LPF) is ubiquitous high pass filters (HPF) can be needed depending on the RF environment or for specific algorithms. The previous post FIR Low Pass Filter Design with Remez demonstrated how to use the remez() function in the SciPy package to design LPF filter weights. The following blog will demonstrate how to use remez() to design a high pass filter as well as designing LPF filters and upconverting them.

Additional examples for Remez filter design can be found in the SciPy documentation under “Examples”.

Design a High Pass Filter with Remez

A HPF can be designed directly with remez() using the following code:

import scipy.signal filterLength = 21 cutoff = 0.375 transBand = 0.1 fPass = cutoff - (transBand/2) fStop = cutoff + (transBand/2) freqVec = [0, fPass, fStop, 0.5] ampVec = [0, 1] HPFRemez = scipy.signal.remez(filterLength,freqVec,ampVec)

The design is similar to that of the LPF discussed in this blog post with the differences being in the cutoff frequency and amplitude vector.

The HPF parameters passed into the remez() function are given in Figure 1, the resulting frequency response in Figure 2 and the magnitude in dB in Figure 3.

Upconvert Low Pass Filter to High Pass Filter

An alternative to designing a HPF directly is to design an LPF and then upconvert the filter weights by multiplication with a sinusoid,

(1) $\begin{equation*}h_{HPF}[n] = h_{LPF}[n] \cdot cos(\phi n).\end{equation*}$

To make the LPF into a HPF it must be upconverted with a center frequency of $f_s/2$ , or in radians $\phi = \pi$ , therefore

(2) $\begin{equation*}cos(\pi n) = (-1)^n,\end{equation*}$

(3) $\begin{equation*}h_{HPF}[n] = h_{LPF}[n] \cdot (-1)^n.\end{equation*}$

Figure 4 gives the impulse response for an LPF and the upconverted HPF.

Frequency Domain Analysis of LPF Upconversion to HPF

The discrete-time Fourier transform of (1) is the frequency-domain convolution of the filter weights $h_{LPF}[n]$ and $\cos(\phi n)$ ,

(4) $\begin{equation*}\begin{split}H_{HPF}\left(e^{j\omega}\right) & = \mathcal{F} \{ h_{LPF}[n] \cdot cos(\phi n) \} \\& = \mathcal{F} \{ h_{LFP}[n]] \} \ast \mathcal{F} \{ cos(\phi n) \}, \\& = H_{LPF}\left(e^{j\omega}\right) \ast \mathcal{F} \{ cos(\phi n) \},\end{split}\end{equation*}$

where $\ast$ is convolution. From (2), conveniently

(5) $\begin{equation*}cos(\pi n) = (-1)^n = e^{j\pi n}.\end{equation*}$

The discrete-time Fourier transform of a sequence $h[n]$ multiplied by a complex exponential is [oppenheim1999, p.59]

(6) $\begin{equation*}\mathcal{F} \{ h[n] \cdot e^{j\phi n} \} = H\left( e^{j(\omega - \phi)} \right).\end{equation*}$

Substituting $\phi = \pi$ , the Fourier transform (4) can then be written

(7) $\begin{equation*}H_{HPF}\left(e^{j\omega}\right) = H_{LPF}\left(e^{j(\omega-\pi)}\right).\end{equation*}$

which is the low-pass filter response frequency shifted to $f_s/2$ or $\pi$ radians. Figure 5 gives the magnitude of the frequency response in (7) and Figure 6 shows the Remez-designed HPF magnitude frequency response is identical to that of the upconverted HPF.

Conclusion

A HPF can be designed directly using the remez() function by passing in the appropriate frequency vector and amplitude vector. Alternatively an LPF can be designed and then upconverted to become a HPF through with

(8) $\begin{equation*}h_{HFP}[n] = h_{LPF}[n] \cdot cos(\pi n) = h_{LPF}[n] \cdot (-1)^n.\end{equation*}$

Have you seen these blogs on filter design?