That's not what TNT said, by his own admission he does not believe in "wasting" time becoming an expert in anything therefore a little sloppiness when making technical comments should be expected.
Oh, there's a simple explanation for that....
Localization is the term used when you use speakers or another external source.
Lateralization is the term used when you have headphones.
If you turn your balance pot and the headphone image moves to either side from inside your head, you are experiencing lateralization. IID based of course, but still..
Jn
Thanks Matt, you showed me those, but rather difficult to use for amp development.
Jacob2,
My read:
Lavry took the NRZ stream, halved the width of the NRZ time, and inserted a zero. So by design, he doubled the sample reconstruction frequency, it's just that half the samples are zero.. Then he shows a spectra where the images fold about 88Khz.
It is consistent with what I would think. If you take a square wave of 50% north, 50% south, and pull back on the on time but keep the rep rate, you are moving towards a stream of pulses and the frequency content changes radically.. So I see that technique as consistent with his graphs around 88Khz as well as stuffing 7 zero values in between the actual data for 8x output. Ya just lose energy in the signal.
jn
And yah, NRZ with zero's is indeed geek humor..
My read:
Lavry took the NRZ stream, halved the width of the NRZ time, and inserted a zero. So by design, he doubled the sample reconstruction frequency, it's just that half the samples are zero.. Then he shows a spectra where the images fold about 88Khz.
It is consistent with what I would think. If you take a square wave of 50% north, 50% south, and pull back on the on time but keep the rep rate, you are moving towards a stream of pulses and the frequency content changes radically.. So I see that technique as consistent with his graphs around 88Khz as well as stuffing 7 zero values in between the actual data for 8x output. Ya just lose energy in the signal.
jn
And yah, NRZ with zero's is indeed geek humor..
This may be an interesting diversion, albeit somewhat of a technical nature.
https://www.diyaudio.com/forums/pow...fit-upgrade-317-based-reg-44.html#post6029370
https://www.diyaudio.com/forums/pow...fit-upgrade-317-based-reg-44.html#post6029370
TNT wrote "Sure - but this article is about persons with severe hearing losses. So phase, as it would be discussed here, is not on the same planet as the article. Your argumentation is often quite sloppy in general. Many of us I suppose notice this - I mention it so that you may adjust as you see fit. You think you educate us but in reality, you rather embarrass yourself."
That is what I responded to. I do not expect RNM to adjust, as that seems to be his style to provoke responses.
We really don't communicate!
No, I was referring to the fact that if wanted all computation could be done these days in 64 bit FP format on a CPU Benchmarks | Intel(R) Integrated Performance Primitives | Intel(R) Software.
As for filtering I still don't understand your need for "proof". The simple pictures show the differences. The filtering in this case is simply convolution with the sinc function which can be demonstrated easily with a single sample and since a digital multiply and add has no source of non-linearity superposition applies. This could be demonstrated in anyone of many math tools and the FFTW library and a tool like BruteFIR could probably run it in real time on a CPU.
Much said here simply violates Nyquist like Wadia's polynomial interpolation, I've learned to expect nonsense from folks that admit they don't understand the principles but not from you.
Yes you don't. I (and others) agree with TNT's assessment you didn't, so what? It's best to not discuss it anymore.
I consider any signal processing done using a digital device to be Digital Signal Processing..I just found it funny you mentioned then independently.
I do not consider extending in time a single sampled value until the next sample arrives to be sinc convolution. Creation of a square wave whose leading edge points are a single data point from a sampled stream creates a truly ugly looking square wave. That process adds energy to the signal, and it does so on the trailing edge of the signal.
My question is indeed very simple. Using hardware, has anybody shown by test, that the NRZ compromise is indeed fully linear. Not sinc convolution, but a true hardware implimentation of NRZ.
I have not stated anything that can be remotely considered as violating nyquist, so would appreciate that inaccuracy be stopped.
At the most, you have misinterpreted my questions, or you have chosen to listen to another with an agenda.
jn
It is About hearing problems but tells what is normal, too. You have to work for it.
I agree it is fast and loose - typing off the top of my head, often.
I dont intend, as I have said many many times, to do more than give you my bottom line conclusions and some clues. It is based upon T&M and listeniing with many systems and gear and often testing for things not part of a spec or text book perfect. For example, I showed these perfect filters you all assume here are the answere... are not good enough at removing HF above the cutoff freq.
In some (many?) cases, i dont have enough details to satisfy those who want to thoroughly get into something. You guys are perfectly capable, it seems.
Go at it. Dont worry about me.
THx-RNMarsh
Last edited:
OMG ! I get that all the time here so now I have whittled down my comments to bottom line only.... Leaving little to argue with except its incompleteness and overly simplistic.
Just an observation that I get same from same all the time. More details seems to elicit more criticism rather than focusing on the point being made or a concept. Often, comments are not even related to anything I said. wow.
You are proof that no level of expertise here matters. Its an ego one-upmanship game for many. Sniping. etal. Boring
-RNM
Last edited:
For scott:
Earlier I mentioned the tradeoff between filter window size and proximity to the nyquist limit. Always above nyquist sampling rate, but close.
Here is a 100 point sampling of a 22Khz signal being sampled at 44.1, and a 500 point sampling. My question in this regard is: without system memory extending into infinity, how can a practical system convert this stream into a constant amplitude 22Khz sine wave?
How can reconstruction algorithms, even 8 or 16 deep IIR's or FIR's reconstruct the flat sine?
The closer I get to nyquist, the longer the sampled stream will take to get to the peak value of the sine.
If I modulate the sine, I can get sampling that duplicates this, so how to reconstruct.
As I back off the frequency, the sample envelope (beat frequency) gets faster, and very soon the peak value is indeed within 8 or 16 samples. And as we go further from nyquist, the actual waveform begins to display exactly. That was Richard's point.
These were my questions.
jn
Earlier I mentioned the tradeoff between filter window size and proximity to the nyquist limit. Always above nyquist sampling rate, but close.
Here is a 100 point sampling of a 22Khz signal being sampled at 44.1, and a 500 point sampling. My question in this regard is: without system memory extending into infinity, how can a practical system convert this stream into a constant amplitude 22Khz sine wave?
How can reconstruction algorithms, even 8 or 16 deep IIR's or FIR's reconstruct the flat sine?
The closer I get to nyquist, the longer the sampled stream will take to get to the peak value of the sine.
If I modulate the sine, I can get sampling that duplicates this, so how to reconstruct.
As I back off the frequency, the sample envelope (beat frequency) gets faster, and very soon the peak value is indeed within 8 or 16 samples. And as we go further from nyquist, the actual waveform begins to display exactly. That was Richard's point.
These were my questions.
jn
Attachments
Thanks, scottjoplin and vacuphile. (I'm more interested by your personal experience than any words of mouth drawn from the literature. Despite i don't agree with the "they measure and sound the same if done right".
Were or are your analog filters for active or passive speakers ?
Last edited:
Whatever I try with headphones, still missing the physical impact on my body, and the fact that I'm afraid a big band is playing inside my head. A very little big band, I agree, but they use to hide their cigarette ends everywhere.
Last edited:
In AM terms, you generate sideband. Reconstruction is possible when the sideband is within Fs/2. But sampling and reconstruction of signal with close to Fs/2 content slam hard against the limitation of practical implementation.
You're welcome. The literature you mention in this case is from other people's experiences, I'm most interested in the commonalities. I don't have the tech to measure everything acoustically, when I set the EQ for the woofers it's by ear, and the time alignment was done by using a tape measure, thankfully not that critical at 200Hz. I've thought about getting all the paraphernalia but it would also mean getting a computer that can run all the software, I really don't want to go down that rabbit hole. I trust my ears enough to get the sound to my liking and have a good idea what's going on, but I don't trust them enough to make a habit of giving my subjective opinions of what I hear, I know how fickle it is from my personal experience. Unfortunately, it also means I'm very dubious when other people give their subjective opinions, I presume they're hearing is similar to mine and everyone else's, ie quite variable and fickle, dependent upon mood, concentration, and any other number of unknown variables.
- Status
- Not open for further replies.