G
Guest
Guest
Are there chips available to generate / measure steps in equal decibels instead of equal microvolts?
Help Support GroupDIY Audio Forum:
SSLtech
Well-known member
No.
It was considered -historically- but logarithmic prediction makes simple DSP tasks MUCH more complicated mathmatically (with the exception of simple gain or loss calculations, which are altered from multiplication so addition... but try and do fourier analysis -as only a SINGLE example- and things get VERY much more complicated....)
Looking from the outside-in, and with simple goals, things would seem simpler, (and level-related distortions become effectively 'companded') but when you try and do 'real' DSP log is a colossal complexity burden, for a comparably insignificant benefit.
In addition, both R/2R and 1-bit converter topologies really don't lend themselves to logarithmic transfer.
CAN it be done? -Yes.
Has it been tried? -Yes.
Was there any significant benefit? -No; only clossal hindrances as soon as you try and move beyond storage/retrieval and simple gain calculations.
Does anyone still do it? -Not so far as I know, but you may which to ask Dan Lavry, Bruno Putzeys or one of the other people to whom I would bow in this field of endeavour.
Keith
Keith
It was considered -historically- but logarithmic prediction makes simple DSP tasks MUCH more complicated mathmatically (with the exception of simple gain or loss calculations, which are altered from multiplication so addition... but try and do fourier analysis -as only a SINGLE example- and things get VERY much more complicated....)
Looking from the outside-in, and with simple goals, things would seem simpler, (and level-related distortions become effectively 'companded') but when you try and do 'real' DSP log is a colossal complexity burden, for a comparably insignificant benefit.
In addition, both R/2R and 1-bit converter topologies really don't lend themselves to logarithmic transfer.
CAN it be done? -Yes.
Has it been tried? -Yes.
Was there any significant benefit? -No; only clossal hindrances as soon as you try and move beyond storage/retrieval and simple gain calculations.
Does anyone still do it? -Not so far as I know, but you may which to ask Dan Lavry, Bruno Putzeys or one of the other people to whom I would bow in this field of endeavour.
Keith
Keith
I'm not sure if I understand your question. The digital data isn't volts or dB, it's bits and you convert it to whatever you want.
Digital information is stored in a powers of two sequence (1,2,4,8,....) so every bit is a 6 db step.
In a simple digital platform it is pretty easy to scale by dB by shifting bits left/right for whole increments of 6 dB, with a multiply for the remaining dB less than 6 using a conversion factor you write into the code.
On my little 8 banger If I want to knock my volume down 6 db that's one bit shift right, If I want to drop the level 3 db I multiply by 0xB5 (approx .707 of full scale 256).
The natural digital data is more like dB than linear, IMO.
JR
Digital information is stored in a powers of two sequence (1,2,4,8,....) so every bit is a 6 db step.
In a simple digital platform it is pretty easy to scale by dB by shifting bits left/right for whole increments of 6 dB, with a multiply for the remaining dB less than 6 using a conversion factor you write into the code.
On my little 8 banger If I want to knock my volume down 6 db that's one bit shift right, If I want to drop the level 3 db I multiply by 0xB5 (approx .707 of full scale 256).
The natural digital data is more like dB than linear, IMO.
JR
SSLtech
Well-known member
Of course, I'm assuming that I have understood the question correctly...
I took it to mean that the 'ladder-step' size increases logarithmically with each step.. i.e. step sizes grow with each successive step. Thus a dB difference would be comprised of a change of an equal number of steps whether the signal is -1dBFS or -71dBFS. Step quantisation error would remain equally significant no matter what the level, and level change would be achieved by addition instead of multiplication.
I am basing my statement that logarithmic prediction was tried and abandoned on some dusty memories of dbx advancing delta-prediction (with compansion) as a viable alternative... they were able to claim unheard of (at the time) dynamic range, but the sheer complexity of doing ANYTHING in DSP afterwards killed the approach stone dead.
Linear is linear. Certain things just work better.
I'm no expert, but I've spoken to a couple about related matters, and while their explanations seemed to make perfect sense at the time, I offer no warranty as to the detail-accuracy of my retelling their stories.
Keith
I took it to mean that the 'ladder-step' size increases logarithmically with each step.. i.e. step sizes grow with each successive step. Thus a dB difference would be comprised of a change of an equal number of steps whether the signal is -1dBFS or -71dBFS. Step quantisation error would remain equally significant no matter what the level, and level change would be achieved by addition instead of multiplication.
I am basing my statement that logarithmic prediction was tried and abandoned on some dusty memories of dbx advancing delta-prediction (with compansion) as a viable alternative... they were able to claim unheard of (at the time) dynamic range, but the sheer complexity of doing ANYTHING in DSP afterwards killed the approach stone dead.
Linear is linear. Certain things just work better.
I'm no expert, but I've spoken to a couple about related matters, and while their explanations seemed to make perfect sense at the time, I offer no warranty as to the detail-accuracy of my retelling their stories.
Keith
X dB is the same number bit shift no matter where it is in the digital word.
Perhaps W. could be more specific regarding his need.
JR
Perhaps W. could be more specific regarding his need.
JR
SSLtech
Well-known member
For gain, yes indeed.
IIRC the original reason for exploring it was in super low-bit days, when the inbuilt 'compansion' result was considered a benefit.
In these 24-bit days, the size of these 'problems' are rather different...
Keith
IIRC the original reason for exploring it was in super low-bit days, when the inbuilt 'compansion' result was considered a benefit.
In these 24-bit days, the size of these 'problems' are rather different...
Keith
G
Guest
Guest
[quote author="SSLtech"]
IIRC the original reason for exploring it was in super low-bit days, when the inbuilt 'compansion' result was considered a benefit.
In these 24-bit days, the size of these 'problems' are rather different...
[/quote]
I'm not going to start a revolution in audio recording; my goal is more prosaic: to trace non-linearity curves directly.
IIRC the original reason for exploring it was in super low-bit days, when the inbuilt 'compansion' result was considered a benefit.
In these 24-bit days, the size of these 'problems' are rather different...
[/quote]
I'm not going to start a revolution in audio recording; my goal is more prosaic: to trace non-linearity curves directly.
I would presume you can do just about anything you want with the data.
For measuring linearity you will be limited to the linearity of the A/D, and will likewise need to scale down voltage being measured to be within input range of A/D (5V or maybe 3.3V p-p). So linearity of that pad/scaling circuitry should be very good.
If using a connected D/A to generate test signal you should be able to perform a calibration, to test your test.
JR
For measuring linearity you will be limited to the linearity of the A/D, and will likewise need to scale down voltage being measured to be within input range of A/D (5V or maybe 3.3V p-p). So linearity of that pad/scaling circuitry should be very good.
If using a connected D/A to generate test signal you should be able to perform a calibration, to test your test.
JR
G
Guest
Guest
One more option is to generate the test signal traditionally like I do now using differential RC network, but measure results using 2 - channel A/D converter instead of comparing them using 2 - ray oscilloscope like I do now.
Similar threads
