Minidsp sampling frequency.

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
My equipment includes an Emotiva UMC-1 Pre/Pro, 2 unbalanced miniDSPs, 1 balanced miniDSP, Emotiva XPA-3, Emotiva UPA-5, Emotiva UPA-2, Behringer EP4000 for 2 subs, etc. No, its not high end but is very good regardless. When someone claims they can hear what science says they can't then I account that to psychology, not to objective fact, unless they have objective test results agreeing with their assertions. In my case what I do and don't hear agrees with the science. I'm certain that if I leave some lighted candles and incense burning while listening it will make the sound more enjoyable and accessible, but this is a psychological effect (which is perfectly valid) not an objective one.
 
ah look up how asyncronous sample rate converters work and see that there is no difference between using 96Khz and 48Khz sample rates from 44.1KHz*. ;) as for number of samples per cycle >=2 fully represents the signal please look up:
Nyquist?Shannon sampling theorem - Wikipedia, the free encyclopedia

rember corectly applied maths dosen't lie but ears do!


*The process applied is an optimisation of upsampling to multi GHz sample frequancies and then downsampling

Nyquist sampling theorem is used for determining the absolute highest input frequency that can be input on an ADC before you get aliasing. For example, if you had a sampling frequency of 44KHz, and you input a frequency that is 30KHz, you would wind up with a lower frequency of something like 10KHz, creating a false frequency waveform. However if you input a triangle or sawtooth wave at 18KHz and your sampling rate is 44 KHz, you will not get a nice looking 18KHz triangular or sawtooth wave in the end.

I disagree with many of the theorems and assumptions that music should just be measured by sinewave. The time domain proponents will rather say that a music signal is a bunch sequential square wave with differening amplitudes, so an ability to reproduce square wave should be more important.

Oon
 
Member
Joined 2007
Paid Member
Nyquist sampling theorem is used for determining the absolute highest input frequency that can be input on an ADC before you get aliasing. For example, if you had a sampling frequency of 44KHz, and you input a frequency that is 30KHz, you would wind up with a lower frequency of something like 10KHz, creating a false frequency waveform. However if you input a triangle or sawtooth wave at 18KHz and your sampling rate is 44 KHz, you will not get a nice looking 18KHz triangular or sawtooth wave in the end.

I disagree with many of the theorems and assumptions that music should just be measured by sinewave. The time domain proponents will rather say that a music signal is a bunch sequential square wave with differening amplitudes, so an ability to reproduce square wave should be more important.

Oon


Incorect the audio Isn't sampled at 48KHz modern ADCs are oversampling therefore the sample rate inside the ADC will be many multiples of this. After sampling decimation is perfomed taking the signal down to the 48Khz sample rate. Therefore all concerns about input components causing aliasing are silly as the input filtering will have effectivly eliminated any components that could cause aliasing.

See also:
Fourier series - Wikipedia, the free encyclopedia
 
Member
Joined 2008
Paid Member
Nyquist sampling theorem is used for determining the absolute highest input frequency that can be input on an ADC before you get aliasing. For example, if you had a sampling frequency of 44KHz, and you input a frequency that is 30KHz, you would wind up with a lower frequency of something like 10KHz, creating a false frequency waveform. However if you input a triangle or sawtooth wave at 18KHz and your sampling rate is 44 KHz, you will not get a nice looking 18KHz triangular or sawtooth wave in the end.

I disagree with many of the theorems and assumptions that music should just be measured by sinewave. The time domain proponents will rather say that a music signal is a bunch sequential square wave with differening amplitudes, so an ability to reproduce square wave should be more important.

Oon

Look, you can only get so much for so little. If you're in pursuit of purity then you should know by now that it doesn't come cheap, so I'm confused what you're expectations are for <$200 DSP. Regardless, if you do a search on Minidsp here and elsewhere you will see that very few if any report any degradation of sound quality. I consider myself very much in the 'everything makes a difference' camp but cannot discern any difference in sound quality with or without my Minidsp - which I consider incredible given the low cost.

Given the subjective affirmations you're looking for (and you are not going to get that from Minidsp, that's not their job) you really have no choice but to buy it and decide for yourself.
 
Last edited:
My point of contention is not about the ADC part. My point of contention is that it is a mistake to think by sampling at 44.1KHz, one can reproduce any random non repetitive waveform accurately (unless of course you particularly enjoy listening to sine wave). Nyquist theorem merely states if you exceeded the maimum allowed frequency you will wind up with rubbish at the lower frequencies and must be removed especially at the ADC part. It doesn't in anyway say that if your frequency is below the maximum sampling frequency, it will be reproduced in accurate high resolution a random input signal. That is a function of of how many datapoints you have. For example, if you have a sampling frequency of 20KHz, and a sampling frequency of 44.1KHz, you only have 2 points for each waveform, how would you know that the original waveform is sine wave, square wave or triangular wave or sawtooth wave?

My first choice would be not to resample at all, and leave the file as it is at 44.1KHz. I definitely agree that resampling the data to 96KHz does not improve on actual data but I would rather have it resampled to 96KHz rather than 48KHz, since with a higher sampling frequency, would mean the interpolated data points are nearer the original datapoints. Resampling to 48KHz on randomly generated waveforms such as music means using the intepolated datapoint rather than the original ones. I would say that would be fine if I had 20 points for a cycle, then it wouldn't make much difference, but when there is only 2 points, then I believe there is a difference. If I am not mistaken, most receivers will actually assume your signal is a sinewave and chooses a sinewave function that actually fits it. For Non oversampling DACs, the believe is this regenation of sinewave actually interferes with the original waveforms, you would actually degrade the signal. The non OS Dac actually performs very poorly in sinewave performance, if you actually loaded a frequency is quite close, you could actually hear the beat of the original frequency and the sampling frequency. If I had my choice I would much rather listen to a 96KHz music, but that is not readily available on most material, so I still have to live with 30 year CD technology.


For some reading on oversampling and DACs, and the actual signal you get out of it.

kusunoki

Oon
 
Member
Joined 2008
Paid Member
I could understand if this was a diy design that needed feedback but the product is done and the Minidsp team seem confident in their design choices. Again, it seems you need to either verify your concerns by listening or go with another option that best fits your criteria for signal integrity.
 
I've already given up the idea of asking for a product change. My debate is mostly with people who are disagreeing with me, and countering me with facts that I disagree with. Like for example there are claims that resampling to 48KHz does not change the sound and it sounds exactly alike, but yet I don't see how this experiment could actually be conducted in the first place. Since it is not possible to run the dsp at 44.1KHz. It is different from saying 48KHz still sounds really good, which I do not dispute.

At no point did I ever say that minidsp sounds like crap. In fact I think it sounds pretty good. I merely said they should not change the original sampling frequency and if they really have to, they should change it to a higher frequency.

Oon
 
Member
Joined 2007
Paid Member
The mini DSP uses this sample rate converter IC:
http://focus.ti.com/lit/ds/symlink/src4382.pdf
it has a SNR of 125dB much better than 16bit audios 96dB and much better than the DAC's used in the mini dsp (93dB?). I am sorry to come accross so harshley but your basic premis that a signals cannot be decomposed is plain wrong, many of the systems I have designed wouldn't work if this was the case.
 
I am not sure what the DAC of the chip is but I don't use it. I am presently using it with the miniamp (from minidsp) for my output. Although I've heard some comments that the DAC is not really that great. The reason I got the minidsp miniamp combo was to try a digital amp, see what happens if everything could go from CD to speaker in a digital form, without passing capacitors, resistors, transitors and volume controls.

While I am familiar quite familiar with the concept of fourier analysis, and that any waveforms can be decomposed into its even and odd harmonics, the concept can only be applied in one direction, a square wave can be decomposed by into its odd hamonics, but the opposite is not true, a fundamental frequency and its odd harmonics do not make a square wave, unless they are perfectyl coherent and the start at the same time. While DSP/processors normally mentions their performance in the frequency domain, there is really little mention on how they actually perform in the time domain. So in general, I disagree on how most systems/chips are actually rated based on sinewave analysis, because it doesn't quite represent the music waveform.

While the datasheet mentions a very high performance. Consider the following, if you have a set of datapoint, how do you determine the datapoints in between them? This is not mentioned exactly in the datasheet. Linear interpolation? polynomial interpolation? Sine wave approximation? What exactly is happening inside? If you look at it this way, I think it would be fair to say the random waveform (music signal) created after resampling to 48KHz would be a bit different than the original at 44.1KHz. Whether one can hear it is another story.

While I I agree that the concept works but to me the question is not whether it sounds good, but rather will it make it sound different?

I think in many ways you and I will have to agree that we are going to disagree on many issues. But that is life...

Oon
 
Member
Joined 2007
Paid Member
By expansion and then decimation. The 44.1Khz data is upsampled to a very high frequancy that is also a multiple of 48Khz and then low pass filtered and decimated. The expansion is just setting the high frequancy samples to the same as the 44.1Khz samples IE is the samples are: [1 2 3], and the expansion factor m is 3 the high frequancy waveform is [1 1 1 2 2 2 3 3 3]. The exact process is different as samples that are not selected by the decimator don't have to be generated but this is mathamaticaly equivlent.

I think your argument is that by taking data into the frequancy domain and then back to the time domain we loose infomation (EG this happens in mp3 encode/decode) and this is correct however this isn't what happens inside a sample rate converter IC or inside a DSP using the kind of filters that the mini-dsp uses they are time domain filters derived from z domain difference equations.

I have a load of slides if you would like to learn some more? I like DSP's alot just CBA with developing my own DSP board and the mini-dsp doses all I want.

You are correct the mini-dsp's DAC's arn't the worlds best:
ADAU1701 | SigmaDSP® 28/56-Bit Audio Processor with Two ADCs and Four DACs | Audio Signal Processors | Audio/Video Products | Analog Devices
but they arn't bad and are have better SNR than the theoretical SNR of 16bit audio.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.