What exactly is DSP?

Digital Signal Processsor.

I am sure a more detailed explanation will be forthcoming from other members.

But in simple terms, DSP takes any signal (audio or video) and applies a mathematical transformation on it to change the outcome. This could be to change the levels, frequencies, colours, whatever. Some parts of the DSP transformation may be hard coded to tune the digital stream according to the specific hardware. You see this with Naim streamers. So the samples reaching the DSP processor get the same transformation again and again - though this can be dependent on the preceding and subsequent samples depending on how clever it is being. Generally, samples are processed in a "window" of a size that varies according to design which takes a certain number of samples at a time to understand the context which helps it determine the appropriate transformation to apply on a specific sample or how to break it into smaller samples (oversampling).

Then there is user definable DSP which you get in AV processors to change the frequency responses according to the room (room correction) and apply a delay signal code to tell the cache how long to hold it for before release.

I don't know what level of detail you are after. There is a whole field of study on DSP and related nomenclature for this which there may be some specialists on the forum.

Sloop John B posted:

What exactly is DSP? (I know what the acronym stands for)

now I'm someone for whom "packets" are what I used to smoke and TCP an antiseptic so please keep it simple.

 

High, as said above DSP stands for Digital Signal Processing. What does it mean?

Well with analogue signals we can pass the signal through a particular circuit to modify it such as filter the signal, edge enhance an analogue video signal  or extract the two stereo channels from a FM stereo broadcast signal. This is called signal processing.

However when the signal is encoded as digital data using a stream of samples or pixels, a different approach is required to achieve the same effect.. Effectively the analogue circuitry functions are mapped to equivalent mathematical functions that are applied to the stream of digital samples. This is called digital signal processing.

Now in our audio world DSP is used to process the digital signals in readiness to sending to the physical digital to analogue encoder. Naim process the digital signal by oversampling it.and low pass filtering it ..this improves the audio performance within the physical DAC chips used.

Additionally realtime DSP in devices is often undertaken within specialised programmable DSP processors. Naim use a SHARC processor. Naim use this DSP processor for other functions as well like reclocking and DSD to PCM conversion.

Simon

the problem is that, without some technical theory, is hard to really understand digital technology...the net is full with approximations of those theories, and thus an entire "sub-culture" of digital has risen... and many wrong or distorted infos have now been published all over...

while the thing is straight enough, and one should read the mathematical theory for digital first... this is not really necessary if one wants to enjoy the sound and goodies of digital tech. but if one needs to really understand, sorry, without math, there will be just a glimpse of it...

but to keep it at the level you implied above, way far off the theory that sustains digital tech, the DSP is merely a processor with dedicated and programmable behaviour. it can be used for either audio or video purposes, and other non-media stuff, as well, fwiw...

Mmmmm, I often pondered the same question, How far off am I if I'm thinking graphically in terms of sin and cos curves/analog vs rectangular boxes/ digital smothed out to to mimick the analog graphs

Allante93

It depends. For audio the analogy is good enough but there is more to it because modern DSP is context aware and not focused on the values at one point in time. For video it moves completely away from the sine/amplitude model you are thinking of.

Worth noting that audio/video is only one part of DSP. Any data set presented as a stream can be processed and when that stream is a representation of a signal you have the possibility of running it through DSP. The uses are huge. Scientific instrument and medical or laboratory devices make heavy use of DSP.

DSP, digital signal processing, is the description given to processing signals which are represented in discretely sampled digital data.  The processing often involves filtering, resampling (changing sample frequency), frequency analysis or a host of other manipulations.  These are accomplished by performing calculations on the data sample values.  

Many applications utilize DSP, which can be quite sophisticated.  Our digital audio devices use DSP in the DAC to effect conversion of digital signals back to analog.  Increasingly we see more applications to digital audio, such as "room correction," crossover filtering and time delay, "pop and click removal," and so forth.  Cell phones are little wonder devices that make sense out of the various radio signals from different towers (I have no idea how...), and spies can separate individual conversations from other sound or noise sources - just to offer a few examples.

Charlie

Didn't Naim use DSP to 'tune' the audio output so that it would sound decent in a car, or more specifically a Bentley?  

A DSP is just a component part of mechanism to perform digital arithmetic on the digital bit-stream. Usually it is used to make filters that are difficult to implement with analogue components after the D to A converter (DAC) before the analogue signal is output to the Pre.

When you sample something you generate things, sampling artifacts, that need to be removed when you put it all back together as an analogue signal. So a DSP is part of what can be used to do that - not all DSPs are the same or used the same, it is just a component of the design.

You can do more than just pure filtering, like 'room-correction' if you have the right mathematical model for absolutely everything critical to do that without introducing non-musical artifacts into the signal - only personal audition will tell you if the design succeeded or not to satisfaction. I've not yet heard room-correction do more good than harm, but I'm sure it can work.

But with the pure filtering operation there is more hope of success, as there are no unknowns, so the design can be implemented knowing all boundary conditions that need to be met - what all engineers really want to have for a project.

DB. 

feeling_zen posted:

Allante93

It depends. For audio the analogy is good enough but there is more to it because modern DSP is context aware and not focused on the values at one point in time. For video it moves completely away from the sine/amplitude model you are thinking of.

Worth noting that audio/video is only one part of DSP. Any data set presented as a stream can be processed and when that stream is a representation of a signal you have the possibility of running it through DSP. The uses are huge. Scientific instrument and medical or laboratory devices make heavy use of DSP.

DSP is simply discrete sample processing, and SP is continuous signal processing.. The former is done arithmetically on digital representation of the analogue signal (IE discrete samples ), the latter is done in the analogue domain such as with electronic components, springs, lenses, servos etc

so to encroach into a bit of theory.. The point above about DSP working on many points in time is key to its success.. It's often called function window or window... The bigger this window the more accurate and less arithmetic distortions are applied to the samples... we sometime see this referred to as taps (derived from old fashioned analogue tape delay processors) .. Therefore simplistically the more taps you have in your filter function or relative points in time, the more accurate the sample processing,.. But the more computing power you need to process each sample from your signal through this large window containing the predefined  sample values representing the filter response at relative points in time. That is the filter function values are multiplied by the digital signal... and this is called convolution... again you sometimes see this term used now in consumer audio DSP.. Especially in room correction software.

i hope that makes some sense.. and ties in some terms banded about in consumer devices.. I loved DSP at Uni in the late 80s and based my final year project on DSP .. Albeit video DSP and still do enjoy it ... if anyone is interested I can recommend some great introductory texts with good worked examples and problems.. 

 Simon