ECE4024 Chap.2 Sampling and Signal Processing
Sampling and Signal Processing
Sampling turns a continuous waveform into a sequence while preserving recoverable information only when spectral replicas remain separated. This chapter develops the frequency-domain picture of aliasing, connects continuous and discrete transforms, and gives a reliable process for rational sample-rate conversion.
The key habit is to run both a rate check and a spectral check: correct arithmetic alone cannot repair overlapping replicas.
What this chapter covers
- 01
Bandlimited signals and spectral support
- 02
Sampling as repeated spectrum replication
- 03
Nyquist rate and anti-alias filtering
- 04
Continuous and discrete frequency descriptions
- 05
DTFT periodicity and DFT bin spacing
- 06
Interpolation images and reconstruction filters
- 07
Decimation and alias protection
- 08
Rational rate conversion with independent checks
Worked example · free
Design a rational sample-rate converter
- +1Reduce the rate ratio: 24/16 = 3/2, so interpolation factor L is 3 and decimation factor M is 2.
- +1Insert two zeros after every original sample. The intermediate rate becomes 16 x 3 = 48 kHz.
- +2Apply a low-pass filter at the intermediate rate to remove interpolation images and protect the later decimator from aliasing.
- +1Retain every second filtered sample, giving 48/2 = 24 kHz.
- +1Check the complete identity: 16 x 3 / 2 = 24 kHz, and confirm the desired baseband lies inside the filter passband.
Key terms
- Nyquist rate
- Twice the highest occupied baseband frequency under the ideal bandlimited reconstruction condition.
- Aliasing
- Irreversible overlap of shifted spectra after sampling or decimation, causing different analogue components to produce the same discrete-time frequency.
- Anti-alias filter
- A filter placed before sampling or decimation to remove frequencies that would overlap after the rate reduction.
- Interpolation
- A sample-rate increase formed by inserting zeros and filtering the resulting spectral images.
- Decimation
- A sample-rate reduction that retains every selected sample after protective low-pass filtering.
- DFT bin
- One sampled frequency point of a finite-length transform, separated from adjacent bins by sampling rate divided by transform length.
Sampling and Signal Processing FAQ
Why must the anti-alias filter come before sampling?
Once high-frequency components fold onto the same discrete-time frequency as desired components, no digital operation can identify their original sources. Filtering first prevents that ambiguity from entering the sample sequence.
Is sampling exactly at twice the bandwidth always practical?
It is the ideal boundary for a perfectly bandlimited signal and ideal filter. Real filters need a transition band, so practical designs normally include sampling margin and specify what out-of-band content is rejected.
What does DFT bin spacing tell me?
It tells you the separation between sampled frequency points and equals the sampling rate divided by transform length. It does not by itself state the signal bandwidth or guarantee that two nearby tones are resolved.
Why does interpolation insert zeros before filtering?
Zero insertion creates samples at the higher grid and produces repeated spectral images. The interpolation filter removes those images and reconstructs the values between original samples on the new time grid.
How do I check a rate conversion answer?
Verify that the input rate multiplied by the interpolation factor and divided by the decimation factor equals the target. Then check that the filter passband keeps the desired spectrum and blocks later aliases.
Exam move
Sketch one baseband spectrum and its nearest replicas until aliasing becomes visually obvious. Practise writing the Nyquist inequality with the exact bandwidth convention stated in the question. For every rational converter, reduce the fraction, write interpolate then filter then decimate, calculate the intermediate and final rates, and name what the filter removes.
When using a DFT, label the sampling rate and transform length before converting a bin index to hertz. Keep analogue anti-alias filtering separate from the digital filter inside a rate converter, because they protect different operations. Organise practice around four pictures: a continuous spectrum, shifted sampling replicas, a discrete-time frequency axis and a rate-conversion block chain.
On each picture, mark the relevant periodicity and the unit on the horizontal axis. This makes it harder to compare hertz with radians per sample or to apply a two-sided bandwidth convention to a one-sided statement. For a sampling question, write the occupied positive-frequency interval before quoting a minimum rate; bandpass signals can require more care than the baseband shortcut.
For a transform question, calculate bin spacing from sample rate divided by transform length, then decide whether the requested index lies in the positive-frequency or wrapped portion. For interpolation by L, insert the conceptual zero samples, identify spectral images and state why the following low-pass filter is needed.
For decimation by M, put the anti-alias filter before the downsampler and show the new sample rate explicitly. In a combined L-over-M converter, record the high intermediate rate even when an efficient polyphase implementation avoids constructing every zero-valued sample. Run unit drills that move among seconds, microseconds, hertz, kilohertz and samples per symbol.
Close each solution with a spectrum check: replicas must not overlap after the protection filter, desired content must remain inside the passband, and the final rate must equal the original rate multiplied by the reduced rational factor.
Working through Sampling and Signal Processing in ECE4024? Sia is AskSia’s AI Electrical Engineering tutor — ask any ECE4024 Sampling and Signal Processing question and get a clear, step-by-step explanation grounded in how ECE4024 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
