L06: Audio Encoding- Quantisation

Question 1

What is the dynamic range?

Answer 1

• Ratio between the largest and smallest differentiable signals

Example: Dynamic Range of Human Hearing
- Min= 0dB
- Max = 140dB
- Dynamic Range = 140 – 0 = 140dB
- A difference in intensity of: 10,000,000 times

Question 2

What is a microphone?

Answer 2

• Used to record audio
• A microphone is a transducer (converts energy from one form to another)
• A microphone converts changes in pressure to induce electrical voltage changes

Question 3

How do microphones work?

Answer 3

Microphones have a diaphragm which moves a coil due to changes in pressure.
We can use a magnets to measure the pressure changes due to the mechanical movement of the coil and convert them into voltage changes (analog).

Question 4

What do we need to do with the readings from a microphone?

Answer 4

• As the voltage changes are analog, we need to sample them
• We must avoid aliasing (sample at least double the maximum frequency of the signal)
• We then need to quantise these samples

Question 5

What happens if the music we are sampling has components above the initially set maximum frequency?

Answer 5

• Any frequency above that will introduce aliasing and will create undesired frequency components
• We need to filter these frequencies before we record them
• We can filter them by sampling at a much higher bandwidth and then apply a low pass filter on the frequencies we dont want (to remove the aliased signal)

Question 6

What do we get when we convert an analog signal into a digital signal?

Answer 6

A series of bit (binary) values
Every x bits will represent a single sample

Question 7

How do we quantise a signal?

Answer 7

We take each sampled value and round it to the nearest unit of measurement for our signal. We want to know the maximum and minimum signals we are getting to use all the available space in the quantisation space.

Question 8

What does quantisation introduce?

Answer 8

Quantisation adds noise
- This is the change in value from the actual sample and the quantised value (which has been rounded)
- Every digital signal that was originally analog will have noise, but it is not necessarily perceivable
- A greater sampling rate means less noise but higher bit rate

Original signal = Quantised signal + Quantisation noise
Quantised signal = Original signal - Quantisation noise

Question 9

What are some sampling standards?

Answer 9

CD quality: 16 bits
-> 2^16 levels to quantise over
Telephone: 8 bits
-> 256 levels to quantise over
-> i.e. 256 levels of loudness

Question 10

What can we hear when we quantise a 16 bit signal down to an 8 bit signal?

Answer 10

• We can hear a hissing noise (quantisation noise)
• Quality decreases

Question 11

How does the change in amplitude affect the quantisation noise?

Answer 11

• The quantisation noise depends on the dynamic range and fluctuation
• The signal to noise ratio is not uniform across amplitudes
• Lower amplitude signals have a lower signal to noise ratio
• Higher amplitude signals have a higher signal to noise ratio

Question 12

How can we maintain a uniform signal to noise ratio?

Answer 12

• To maintain a uniform signal to noise ratio, we can compensate by increasing the quantisation levels for low amplitudes or changes in amplitudes (flat signals), and decreasing the quantisation levels for high amplitudes and high changes in amplitudes (steep signals).
• We no longer want to spread the quantised values out equally -> depends on amplitude
• There is a higher density of quantisation steps at lower amplitudes and a lower density at higher amplitudes above.
• This is a type of compression- we compress the mapping of the original signal to an encoded signal.

Question 13

What is companding?

Answer 13

Compression + Expanding

These curves tell us how to convert a linear signal into an encoded signal. Tells us how to differentiate signals and compress them such that we have more steps and reduced quantisation noise for lower amplitude signals. This will give us more of a uniform quantisation noise at all different amplitude levels.

Changes the relative quantisation noise for each step, decreasing the signal to noise ratio for lower amplitudes (but potentially increasing it for higher amplitudes)

There are standards for this compression
- Algorithms: A-law (Europe), μ-law (US, Japan)

Question 14

What are the sub-processes in the conversion of an analog signal to a digital signal?

Answer 14

Capture
Convert
1. Filter
2. Sample
3. Quantise
Process

Question 15

What can you not avoid when you process a signal?

Answer 15

Quantised noise
We can reduce it by increasing the bits/sample