Applications of Filters in Signal Processing | ECExplain

Visual

Concept overview

A filter selectively passes or attenuates signals based on their frequency. The four basic types are low-pass (passes frequencies below a cutoff), high-pass (passes frequencies above a cutoff), band-pass (passes a range of frequencies), and band-stop or notch (attenuates a narrow frequency band). Filters are implemented as passive circuits using R, L, and C components, as active circuits using op-amps with RC networks, or as digital FIR and IIR filters running on DSP processors and FPGAs. The choice of Butterworth, Chebyshev, or Bessel approximation determines the trade-off between passband flatness, roll-off steepness, and phase linearity.

Real-world applications

FIR Low-Pass Filter in TMS320C6748 DSP Audio Processor

Professional Audio / Consumer Electronics

In professional audio equipment such as digital mixing consoles and studio interfaces, a symmetric FIR low-pass filter running on Texas Instruments TMS320C6748 or similar DSPs removes out-of-band noise above 20 kHz before the digital-to-analog converter output stage. A linear phase FIR filter is chosen over IIR because it introduces constant group delay across the audio band, preserving transient accuracy in musical content.

SAW Band-Pass Filter in 4G LTE RF Front End

Telecommunications / RF Engineering

Surface acoustic wave (SAW) filters from Murata and TDK are placed between the antenna and the low-noise amplifier in LTE smartphones. The SAW filter passes only the specific 20 MHz or 40 MHz LTE band allocated by spectrum regulators while rejecting signals from adjacent channels, Wi-Fi, and other cellular bands. This prevents the LNA from being saturated by out-of-band interference.

60 Hz Notch Filter in ECG Signal Chain

Biomedical Electronics / Healthcare

ECG acquisition systems embed a 60 Hz (or 50 Hz in India) IIR notch filter in the digital signal chain to remove power line interference coupled capacitively through the patient's body and electrode cables. A second-order IIR notch using a pair of complex conjugate zeros on the unit circle at 60 Hz removes the interference while introducing minimal distortion to the clinical ECG waveform frequencies between 0.05 Hz and 150 Hz.

How it works in practice

In an active Sallen-Key second-order low-pass filter, an op-amp buffers the output of a passive RC ladder network. The feedback from the op-amp output to the mid-node of the RC ladder provides frequency-selective positive feedback, which sharpens the roll-off beyond what a passive RC achieves. The quality factor Q and the natural frequency w0 are set by the four passive components R1, R2, C1, C2. In a digital FIR filter, the input signal is convolved with a set of fixed tap coefficients h[0], h[1], ..., h[N-1]. Each output sample y[n] = sum of h[k] * x[n-k] for k from 0 to N-1. Increasing the number of taps N sharpens the transition band but increases computational load linearly. An IIR filter uses feedback: y[n] depends on both previous inputs and previous outputs, giving a much steeper roll-off per coefficient but introducing nonlinear phase and potential instability if coefficients are not carefully computed. Bilinear transformation is the standard method for converting an analog prototype filter such as a Butterworth or Chebyshev design into a stable IIR digital filter.

Examples

Anti-Aliasing Filter for 10 kHz ADC Sampling

An ADC sampling at 10 kHz must reject all signal energy above 5 kHz to prevent aliasing. A fourth-order Butterworth active low-pass filter with cutoff at 4 kHz is placed before the ADC input. Using two cascaded Sallen-Key stages, each second-order stage requires two op-amps such as the MCP6002 and four passive components. The resulting attenuation at 5 kHz is approximately 40 dB, adequate for most sensor data acquisition applications.

LC Band-Pass Filter in AM Radio IF Stage

In a superheterodyne AM radio receiver, the intermediate frequency (IF) stage operates at 455 kHz. A double-tuned LC band-pass filter selects the 10 kHz bandwidth centred on 455 kHz and rejects the image frequency and adjacent AM channels. The two coupled LC tanks are wound on ferrite cores and trimmed with variable capacitors during manufacturing. The filter's selectivity, expressed as Q, determines the receiver's ability to separate adjacent channels 10 kHz apart.

Future scope

Reconfigurable analog filter banks implemented in bulk acoustic wave (BAW) resonator technology are being developed for 5G mmWave handsets, where the filter must switch between dozens of frequency bands rapidly. Companies such as Akoustis Technologies are fabricating BAW filters on silicon to achieve insertion losses below 1 dB at 28 GHz. In machine learning-based audio processing, learned filterbanks in convolutional neural networks trained end-to-end on raw audio waveforms are replacing hand-designed mel-scale filter banks for speech recognition, with research from Google and Baidu showing improved robustness to environmental noise in keyword spotting tasks.