Signal processing is an electrical engineering subfield that focuses on analysing, modifying and synthesizing signals such as sound, images and biological measurements. Electronic signal processing was first revolutionized by the MOSFET and then single-chip digital signal processor (DSP). Digital signal processing is the processing of digitized discrete-time sampled signals. Processing is done by general-purpose computers or by digital circuits such as ASICs, field-programmable gate arrays or specialized digital signal processors (DSP chips).
Hackaday has published an in interesting series of articles on signal processing, and here are some picks from it:
RTFM: ADCs And DACs
https://hackaday.com/2019/10/16/rtfm-adcs-and-dacs/
DSP Spreadsheet: IQ Diagrams
https://hackaday.com/2019/11/15/dsp-spreadsheet-iq-diagrams/
Sensor Filters For Coders
https://hackaday.com/2019/09/06/sensor-filters-for-coders/
DSP Spreadsheet: FIR Filtering
https://hackaday.com/2019/10/03/dsp-spreadsheet-fir-filtering/
Fourier Explained: [3Blue1Brown] Style!
https://hackaday.com/2019/07/13/fourier-explained-3blue1brown-style/
DSP Spreadsheet: Frequency Mixing
https://hackaday.com/2019/11/01/dsp-spreadsheet-frequency-mixing/
Spice With A Sound Card
https://hackaday.com/2019/07/03/spice-with-a-sound-card/
- check also A real-time netlist based audio circuit plugin at https://github.com/thadeuluiz/RTspice
Reverse Engineering The Sound Blaster
https://hackaday.com/2019/06/19/reverse-engineering-the-sound-blaster/
FM Signal Detection The Pulse-Counting Way
https://hackaday.com/2019/08/28/fm-signal-detection-the-pulse-counting-way/
DSP Spreadsheet: IQ Diagrams< https://hackaday.com/2019/11/15/dsp-spreadsheet-iq-diagrams/
Here is an extra, not from Hackaday, but an interesting on-line signal processing tool for generating sounds
https://z.musictools.live/#95
172 Comments
Tomi Engdahl says:
https://hackaday.com/2024/05/16/is-the-frequency-domain-a-real-place/
Tomi Engdahl says:
https://www.digitaldjtips.com/over-to-you-when-is-it-ok-to-go-into-the-red/
Tomi Engdahl says:
https://hackaday.com/2024/06/05/fourier-the-animated-series/
Tomi Engdahl says:
https://hackaday.com/2024/06/23/fixed-point-math-exposed/
Tomi Engdahl says:
Encoding sound in the cochlea: from receptor potential to afferent discharge
https://physoc.onlinelibrary.wiley.com/doi/full/10.1113/JP279189?fbclid=IwZXh0bgNhZW0CMTEAAR0wwGVES_9S9c48HD3VS52LkSU9PqdBSvuBq40Grp3F0Qqm8V1FK9-xy3A_aem_4ucA-ly_KkLCwS5q75TwKw
Tomi Engdahl says:
https://github.com/jaakkopasanen/Impulcifer
Tomi Engdahl says:
https://hackaday.com/2024/10/12/approximating-an-adc-with-successive-approximation/
Tomi Engdahl says:
Reverse engineering the circuit that revolutionized delay effects
https://www.youtube.com/watch?v=4LjP5Y1yxXs
Here’s an interesting problem: how do you create a delayed duplicate of a sound without recording it to some sort of storage medium? Back in the days before digital signal processing and cheap, abundant memory, this was a prime engineering issue. Until two Engineers named Sangster and Teer came up with a deceptively simple solution: the bucket brigade delay.
In this video, I attempt to reverse engineer the architecture of a classic BBD, recreate a bare bones version on the breadboard – and then use a proper BBD chip to design a simple DIY audio delay. If you want to build along, here’s the bill of materials:
Tomi Engdahl says:
Most people don’t actually understand what the differences are in bit rates and Sample rates
Your bit rate is an expression of the noise floor
Zero is always zero but the more bits you add the further down and the noise floor you get
So it’s 16 bit you have -96 DB of noise floor
Every bit gains you another -6 DB of noise floor
Once you get below the noise floor of the equipment the bits no longer matter and at that point it just becomes marketing
What is super useful is to use a floating point
If it isn’t floating point then the extra bits won’t matter because you have more self noise in the equipment than the bitrate you are using
So if you have a 32-bit floating point you can take advantage of every single bit in the mix down that is above the noise floor and it allows you to change the clipping point so that you can mix without having to worry about distortion and then when you did their things down you can essentially select where that noise floor ends up at
And all of this is a moot point if you don’t have software that takes advantage of the floating point to dynamically move the clipping point of your content on the Fly
In live audio it just doesn’t matter as your typical noise floor is more in the -70 range at best
The biggest reason to have a 32-bit floating point bit rate is so that when you dither the product back down to 16 bit or 24 bit that you can get rid of the noise floor created by the objects use to record and throw away those bottom eight or 16 bits of noise allowing for a cleaner and less noisy end product
Where your best and biggest differences you can hear happen are in the sample rate
And once you get above 96k there are diminishing returns but you can start to hear a smoothness and the sound with more sample rate
In the critical listening side of things once you get to 384 k sample rate it becomes indistinguishable from analog circuitry as the noise created from each sample becomes quieter than the noise of any analog circuit so you are left with only noise created from The Analog Devices connected to the converter
Tomi Engdahl says:
Why Your 16-bit CDs Might Sound Better Than 24-bit Hi-Res Streaming
https://www.headphonesty.com/2024/11/cds-offer-sound-quality-high-res-streaming/
Tomi Engdahl says:
Concise summary:
Bit Depth: Determines the noise floor in digital audio. Each bit adds -6 dB to the noise floor. For instance, 16-bit audio has a -96 dB noise floor. However, once below the noise floor of the recording equipment, extra bits have no practical effect and are mostly marketing unless using floating point.
Floating Point: A 32-bit floating point system is highly advantageous for mixing because it allows dynamic clipping point adjustments and better utilization of bits above the noise floor. It also ensures cleaner output when dithering back down to 16-bit or 24-bit, as it can eliminate noise introduced by recording equipment.
Sample Rate: Affects the smoothness of the audio. Differences are noticeable as you move above 96 kHz, but returns diminish beyond that. At 384 kHz, the sound becomes almost indistinguishable from analog due to reduced sample noise.
Practical Application: In live audio, bit depth and floating point often don’t matter much due to higher inherent noise floors (-70 dB). The real benefits of high bit depth and sample rates are more relevant in critical listening and post-production.
Tomi Engdahl says:
https://hackaday.com/2024/11/17/analog-shift-register-revealed/
Tomi Engdahl says:
Trivia: The main reason we have 48 kHz sampling (and silly multiples of it ) is that it works with 24 frames/second film. 44.1 kHz does not provide a whole number of samples per picture frame. 48 kHz of course also works well with 50 and 60 frames/s
Tomi Engdahl says:
If your DAC is 32 bit and your source is 24 bit, you can lower (in amplitude) the source/input by 8 bit/48 dB without loosing any bits. Comes in handy if your DAC hasn’t a hardware/analog volume knob.
Tomi Engdahl says:
32bit (float) is useful (kinda) for recording out in the wild, like for documentaries or maybe even movies (although it’s less helpful there since you have sound levels that can usually be anticipated and also a dedicated person for mixing and gain staging).
It basically eliminates the need to adjust the need to adjust your gain level between loud and silent takes.
In post production you can then adjust the levels without raising the noise floor (into an audible level)
Why ever you would need 32bit audio on the end-user side of production is completely beyond me.
Tomi Engdahl says:
a normal person’s “overkill” is an audiophool’s gotta have.
A standard single-precision floating point value has only a 23-bit mantissa.
The reason such things make sense is that they are so trivial to implement that smaller word size/integer implementations are pointless; chips with more-hefty circuits are trivial to fabricate, and high-precision/high-dynamic range claims are good for product marketing.
You need a good deal of precision/dynamic range when you’re processing audio (i.e., creation). For playback — not so much.
Tomi Engdahl says:
Here’s Why 32-bit/384 kHz DACs Make Little Sense
https://www.headphonesty.com/2024/01/32-bit-384-khz-dacs-make-little-sense/
Tomi Engdahl says:
https://www.edn.com/earplugs-ready-lets-make-some-noise/#google_vignette
Tomi Engdahl says:
Recording uses 32-bit *floating point*, for playback it is of course completely unnecessary.
See https://tascam.jp/int/feature/32-bit_float
Tomi Engdahl says:
The Infrasonic 17 Hz Tone Experiment Induced Revulsion And Fear In Participants
Concert-goers were subjected to unusual sounds, with unsettling results.
https://www.iflscience.com/the-infrasonic-17-hz-tone-experiment-induced-revulsion-and-fear-in-participants-77093