Sampling Rate Conversion

Sampling rate conversion is the process of converting a discrete-time signal from one sampling rate to another without passing through the analog domain. When the ratio of output rate to input rate is a rational number expressible as L/M where L and M are positive integers, the process is called rational factor sampling rate conversion. This topic is central to multirate DSP and appears directly in GATE questions related to digital filter design and multirate systems.

Core Concept: Rational Factor Conversion

Sampling rate conversion is required whenever two digital systems operating at different rates must exchange data. If the ratio of desired output rate to input rate can be expressed as a rational number L/M where both L and M are positive integers, conversion is achieved using a cascade of an upsampler by L, a combined lowpass filter, and a downsampler by M. The order is critical: upsampling must always precede downsampling in this structure.

The reason upsampling must come first is to prevent aliasing. If downsampling by M is performed first and the signal bandwidth exceeds pi/M, aliasing occurs and the information is permanently destroyed. By upsampling first, the signal is moved to a higher rate, and the combined lowpass filter can then suppress both the images (from upsampling) and any potential aliases (from subsequent downsampling) before the decimation stage.

An important computational advantage is that the two filters required in a naive implementation (anti-imaging LPF for interpolation and anti-aliasing LPF for decimation) can be merged into a single combined LPF operating at the high intermediate rate L.Fs. This reduces implementation complexity significantly, particularly when combined with polyphase decomposition.

Mathematical Expression

The cutoff frequency of the combined lowpass filter used in L/M rate conversion is:

wc = min(pi/L, pi/M)

This ensures the filter simultaneously prevents imaging (by having cutoff no greater than pi/L) and prevents aliasing (by having cutoff no greater than pi/M). The passband gain is set to L to compensate for the amplitude reduction from zero-insertion during upsampling. The output sampling rate is:

Fs_out = (L/M) x Fs_in

When L and M share a common factor, they should be reduced to their coprime form (GCD removed) before implementing the system to minimize computational load. For example, converting from 32 kHz to 48 kHz gives L/M = 48/32 = 3/2 in reduced form.

Practical Understanding

Rational sampling rate conversion is ubiquitous in modern audio and communication systems. The conversion between CD audio (44.1 kHz) and professional studio audio (48 kHz) is a widely cited example. The ratio 48/44.1 simplifies to 160/147, meaning the system must upsample by 160 and downsample by 147. Although this seems computationally expensive, polyphase filter structures make it practical by distributing filter computation efficiently across subfilters.

In software-defined radio, signals arriving from different bandwidth channels must be brought to a common processing rate. Rational rate converters allow flexible decimation and interpolation to be applied in a single pipeline. The combined LPF in such systems is typically a linear-phase FIR filter, chosen because it introduces symmetric delay and no phase distortion within its passband.

For GATE aspirants, the most commonly tested scenario is finding the LPF cutoff for a given L/M ratio and identifying the output sampling rate. It is also important to remember that even if L equals M (unity rate conversion), the intermediate filtering step is still meaningful when the purpose is to reshape the spectrum.

Given:
Input sampling rate Fs_in = 44100 Hz (CD audio)
Desired output rate Fs_out = 48000 Hz

Why this formula applies:
Rational factor conversion uses L/M where L/M = Fs_out / Fs_in.
The combined LPF cutoff = min(pi/L, pi/M) prevents both imaging and aliasing.

Formula:
L/M = Fs_out / Fs_in (reduced to coprime integers)
wc = min(pi/L, pi/M)
Fs_out_actual = (L/M) x Fs_in

Substitution:
L/M = 48000 / 44100 = 160/147 (GCD = 300)
L = 160, M = 147
wc = min(pi/160, pi/147)

Calculation:
pi/160 = 0.00625*pi rad/sample
pi/147 = 0.006803*pi rad/sample
min = pi/160 = 0.00625*pi rad/sample

Final Answer:
Upsample by L = 160
Downsample by M = 147
Combined LPF cutoff = pi/160 rad/sample (at intermediate rate 160 x 44100 = 7,056,000 Hz)
LPF gain in passband = 160
Output rate = (160/147) x 44100 = 48000 Hz confirmed

Exam Tip: Always reduce L/M to coprime integers before computing LPF cutoff. The combined filter cutoff is min(pi/L, pi/M) not pi/(L+M). If L > M, the cutoff is pi/L and gain = L. If M > L, the cutoff is pi/M and gain = L still.

Quick Revision

• Rational rate conversion by L/M: upsample by L, apply combined LPF, downsample by M. Order is fixed; upsample always comes first.
• Combined LPF cutoff = min(pi/L, pi/M). Passband gain = L always.
• L and M must be coprime (reduce GCD before implementation) for minimum computational cost.
• Output rate = (L/M) x Input rate. Real example: 44.1 kHz to 48 kHz needs L=160, M=147.
• Single combined LPF replaces two separate filters (anti-imaging + anti-aliasing), operating at the intermediate rate L x Fs.
• Exam trap: interchanging L and M, or computing LPF cutoff as pi/M when L is larger. Always take the minimum of pi/L and pi/M.
• The intermediate sample rate is L x Fs_in. Polyphase decomposition reduces actual computation to near the output rate.