My apologies for being long-winded in my previous reply; I forgot to comment on this part. It's incorrect to say that dither only makes the lowest bits linear and not the rest. A 16-bit sample is a whole unit. Dither makes the whole unit contain the sample without input-dependent distortion and noise modulation. There really is no way to affect only the lowest bits (except for truncation, or other forms of masking, which adds distortion to the unit represented by the remaining bits). In other words, proper dither makes all bits linear.
I suppose it's potentially confusing when TPDF of 2 LSB (amplitude) is discussed, because that might be construed as affecting only the 2 least significant bits, but that is not the case. Adding anything to a sample, positive or negative, can and usually does affect all bits.
By the way, when the TPDF amplitude of 2 LSB is mentioned in JAES v40 #5, it is explained that only this amplitude is correct. Less than 2 LSB peak-to-peak, or more than 2 LSB peak-to-peak is not successful in eliminating input-dependent distortion and noise modulation. Thus, discussions of ever-increasing amounts of dither being applied are not consistent with the literature that I happen to have referenced.
Hi,
ASRC is used whenever it is not possible to establish an integer ratio between two sample rates. For example a nominal 192KHz sample rate device will have a REAL sample frequency determined by it's clock that is not exactly 192.0 period KHz.
Moreover, the target device may operate not precisely at 44.1KHz. In this case the fixed 147 times upsampling and 640 times downsampling will cause problems.
So you have to somehow re-clock the whole shebang, which is in effect offered in your case by a very large fifo with independent write and read clocks... (BTW, the FIFO is of course the fie stored on the HDD, it must be written first and can be read back independently - classic FIFO behaviour).
Ciao T
ASRC is used whenever it is not possible to establish an integer ratio between two sample rates. For example a nominal 192KHz sample rate device will have a REAL sample frequency determined by it's clock that is not exactly 192.0 period KHz.
Moreover, the target device may operate not precisely at 44.1KHz. In this case the fixed 147 times upsampling and 640 times downsampling will cause problems.
So you have to somehow re-clock the whole shebang, which is in effect offered in your case by a very large fifo with independent write and read clocks... (BTW, the FIFO is of course the fie stored on the HDD, it must be written first and can be read back independently - classic FIFO behaviour).
Ciao T
Hi,
Marantz CD-12/DA-12 or the equivalent Philips LHH1000 combo, fully restored and modernised (capacitors, op-amp's etc.) if we are talking about essentially "stock".
Otherwise any "high end" japenese multibit player, really well modified and with tube output stage, players with TDA1541 also converted to Non-Os are preferred, others exist, AD1862 & PDM100 equipped units probably come next in list.
Ciao T
Marantz CD-12/DA-12 or the equivalent Philips LHH1000 combo, fully restored and modernised (capacitors, op-amp's etc.) if we are talking about essentially "stock".
Otherwise any "high end" japenese multibit player, really well modified and with tube output stage, players with TDA1541 also converted to Non-Os are preferred, others exist, AD1862 & PDM100 equipped units probably come next in list.
Ciao T
Hi,
You mean Dustin?
Not surprised at all...
Ciao T
No it was Martin M. I was having a drink with the CEO of ESS, Martin, and a third person I did not know. They did use "we" when describing the process, and begged me to come listen but I didn't have the time. It was Martin's expertise at ADI, but I see (I think) he is off making finished DAC's now. They were very enthusiastic about using the listening process in the design, and the FPGA eliminates a silicon rev for every tweek or experiment. Yes, it gives me pause when the priests leave the order.
FWIW, I did the op-amp (partly) in the 1862, that's when I cooked up that altenative solution.http://www.cmoset.com/uploads/6A.4-08.pdf
Last edited:
Digital Recording
I appreciate you taking the time to remind me of my mis-use of the terminology, and for sure I realize this. And it is true that dither is input to the system so it adds to the signal, potentially modulating the highest bit, should the signal level be less than 1/2p-p of the dither amplitude away from it (note to self: accuracy of thought is inversely proportional to IPA consumption).
My poorly stated comment was meant to express how the concept of 16-bit sampling in reality involved far fewer bits for most audio. (caution, old fart memory warning: In the early days of digital cassette replication we designed and built a digital master bin. Since the replication ratio was 64:1, the D>A had to handle 20kHz * 64 or 1.28MHz. In 1987 a 16-bit DAC which could handle this was a severely expensive item. As a result we spent a lot of time listening to far less expensive 12 and 14-bit DACs to see if we could hear any compromise in the sound quality...and we sure could. It is definitely true that even with 16-bit systems, level setting is critical...
I may be chasing ghosts here, but let me ask my question using a diffferent thought experiment: Early on DBX touted that they could acheive equivalent 24-bit performance by linear companding wrapped around a 16-bit digital recording system. This is such a loaded statement there is a lot to address, but the idea really bothered me, and my objection can be illustrated by taking the idea to extremes: Assuming a lossless compander (if such a thing could ever exist). Could you actually recover the inputted audio after a 2-bit (w/dither of course) digital encode>decode cycle if you compressed the audio to not exceed the dynamic range of 2-bits?
Also, can one of you who has a handle of the intricasies of digital sampling theory speak to the issue of narrow-band distortion at sub-harmonics of the sampling rate? I noticed this phenomena myself during testing of the aforementioned system.
Thanks for your time,
Howard Hoyt
CE - WXYC-FM
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
I appreciate you taking the time to remind me of my mis-use of the terminology, and for sure I realize this. And it is true that dither is input to the system so it adds to the signal, potentially modulating the highest bit, should the signal level be less than 1/2p-p of the dither amplitude away from it (note to self: accuracy of thought is inversely proportional to IPA consumption).
My poorly stated comment was meant to express how the concept of 16-bit sampling in reality involved far fewer bits for most audio. (caution, old fart memory warning
: In the early days of digital cassette replication we designed and built a digital master bin. Since the replication ratio was 64:1, the D>A had to handle 20kHz * 64 or 1.28MHz. In 1987 a 16-bit DAC which could handle this was a severely expensive item. As a result we spent a lot of time listening to far less expensive 12 and 14-bit DACs to see if we could hear any compromise in the sound quality...and we sure could. It is definitely true that even with 16-bit systems, level setting is critical...I may be chasing ghosts here, but let me ask my question using a diffferent thought experiment: Early on DBX touted that they could acheive equivalent 24-bit performance by linear companding wrapped around a 16-bit digital recording system. This is such a loaded statement there is a lot to address, but the idea really bothered me, and my objection can be illustrated by taking the idea to extremes: Assuming a lossless compander (if such a thing could ever exist). Could you actually recover the inputted audio after a 2-bit (w/dither of course) digital encode>decode cycle if you compressed the audio to not exceed the dynamic range of 2-bits?
Also, can one of you who has a handle of the intricasies of digital sampling theory speak to the issue of narrow-band distortion at sub-harmonics of the sampling rate? I noticed this phenomena myself during testing of the aforementioned system.
Thanks for your time,
Howard Hoyt
CE - WXYC-FM
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
Thorsten,
Here we are going to disagree. If you were using an IEC standard sound level meter even on fast weighting and "C" scale there are issues not covered.
Doing live symphony reinforcement in a pavilion setting, I find that at 30 db of headroom I cannot hear the sound system clip. At the commonly accepted 20 db, careful listening shows the limits. 10 db used to be the standard and that is bad enough I wouldn't use that for a subway station.
The next issue is lower level. When the low SPL levels are shown you again need 30 db below that or you will hear the process noise!
So before looking at the orchestra's range you start with a 60 db guard band.
Now my high reading at the conductors position is 112 db during a rehearsal. The low reading should be the concert halls background noise level of around 25 dba.
ES
Hi,
I used a meter set to fast which had it's "C" setting replaced by a "flat" one
Well, for system clipping I would check the Amp's monitors. Also, remember that you likely use close miking, while me listening (and bootlegging) is already in the far field.
Speaking of sound reinforcement for symphonic music.
I used to regularly go to the summer concerts at Kenwood House at Hamstead Heath in London. Entrance to the park was free and if you did not mind sitting a bit away from the orchestra, so was the music. It was a lovely outing almost every weekend in the summer.
Some concerts I did buy tickets and got to rub shoulders with people rich, famous or monied and blessed with taste (much less rare in England than in the former colonies). The orchestra played in a lovely late 18th or early 19th century Shell type bandstand. No sound reinforcement. Some times it rained.
One ticket I always bought was the finale of the years series. It featured 1812 including her royal majesty the Queen's Royal Horse Artillery with 13 pounder cannons and muskets (Telarc, eat your heart out).
I stopped going when they moved the concerts to a different area, with a stage and sound reinforcement, even though tickets became cheaper and many companies handed out incredible freebies (like free bottle of Cointreu) in the name of promotion. I spend more at the Royal Festival Hall afterwards. Great hall, no PA.
Past that, I analysed uncompressed recordings using software. It is possible that the recordings contained clipped peaks, but this was not audible. I do however not remember seeing any "overs" when recording them (portable DAT and concident crossed 8 ribbon mike - not very stealthy, but I got away with it).
I am unsure if I agree on the 30dB for the foot room and headroom, but if we wish, these are things we can debate AND measure...
Well, recording analogue we had around 80db total usable dynamic available. The orchestra has normally > 50dB between pppp and ffff. I do not remember using 30dB gain-riding in those days (quite frankly, 30dB gainriding is an impossibility even using the Eckmiller W88 fader or it's east german equivalent (meaning frequency compensated so subjective tonal balance did not change when "gain-riding), using more traditional faders I doubt I'd comfortably go much past 10dB...
But we managed... With magnetic tape... Somehow.
We also managed later with 16/48.
Not that I am against 32/384, as long as we get real 20 Bit converters, not 9.5 Bits in hardware and 22.5 Bit in handwaving, marketing and magic...
I do agree with either number, for an empty hall and the conductors rostrum from my "formal" recording career.
Mine where given for centre 4th to 6th row and with a packed audience.
So we are comparing Apples and Oranges.
Ciao T
I used a meter set to fast which had it's "C" setting replaced by a "flat" one
Well, for system clipping I would check the Amp's monitors. Also, remember that you likely use close miking, while me listening (and bootlegging) is already in the far field.
Speaking of sound reinforcement for symphonic music.
I used to regularly go to the summer concerts at Kenwood House at Hamstead Heath in London. Entrance to the park was free and if you did not mind sitting a bit away from the orchestra, so was the music. It was a lovely outing almost every weekend in the summer.
Some concerts I did buy tickets and got to rub shoulders with people rich, famous or monied and blessed with taste (much less rare in England than in the former colonies). The orchestra played in a lovely late 18th or early 19th century Shell type bandstand. No sound reinforcement. Some times it rained.
One ticket I always bought was the finale of the years series. It featured 1812 including her royal majesty the Queen's Royal Horse Artillery with 13 pounder cannons and muskets (Telarc, eat your heart out).
I stopped going when they moved the concerts to a different area, with a stage and sound reinforcement, even though tickets became cheaper and many companies handed out incredible freebies (like free bottle of Cointreu) in the name of promotion. I spend more at the Royal Festival Hall afterwards. Great hall, no PA.
Past that, I analysed uncompressed recordings using software. It is possible that the recordings contained clipped peaks, but this was not audible. I do however not remember seeing any "overs" when recording them (portable DAT and concident crossed 8 ribbon mike - not very stealthy, but I got away with it).
I am unsure if I agree on the 30dB for the foot room and headroom, but if we wish, these are things we can debate AND measure...
Well, recording analogue we had around 80db total usable dynamic available. The orchestra has normally > 50dB between pppp and ffff. I do not remember using 30dB gain-riding in those days (quite frankly, 30dB gainriding is an impossibility even using the Eckmiller W88 fader or it's east german equivalent (meaning frequency compensated so subjective tonal balance did not change when "gain-riding), using more traditional faders I doubt I'd comfortably go much past 10dB...
But we managed... With magnetic tape... Somehow.
We also managed later with 16/48.
Not that I am against 32/384, as long as we get real 20 Bit converters, not 9.5 Bits in hardware and 22.5 Bit in handwaving, marketing and magic...
I do agree with either number, for an empty hall and the conductors rostrum from my "formal" recording career.
Mine where given for centre 4th to 6th row and with a packed audience.
So we are comparing Apples and Oranges.
Ciao T
Last edited:
Scott,
AD1862 & PDM100, NOT AD1855 (I heard that part, I'd take cirrus logic instead any day, which is about the same as saying I'd rather eat at MickeyD). It is not the manufacturer, it is the specific chip.
My personal list for 16/44 goes (either Non-Os or PDM100 for HDCD):
#1 TDA1541AS2
#2 TDA1541AS1
#3 TDA1541A
#4 AD1862
#5 PCM56
#6 PCM63
#7 and on - the rest of multibitdom
Listening to TDA1541AS2 with JAN GE 6072A ***** as analogue stage ATM...
Ciao T
AD1862 & PDM100, NOT AD1855 (I heard that part, I'd take cirrus logic instead any day, which is about the same as saying I'd rather eat at MickeyD). It is not the manufacturer, it is the specific chip.
My personal list for 16/44 goes (either Non-Os or PDM100 for HDCD):
#1 TDA1541AS2
#2 TDA1541AS1
#3 TDA1541A
#4 AD1862
#5 PCM56
#6 PCM63
#7 and on - the rest of multibitdom
Listening to TDA1541AS2 with JAN GE 6072A ***** as analogue stage ATM...
Ciao T
Yes. 1.5 uSec has been discerned, lateralization.
Why? This is a non-sequitur. (this unit must survive)
It is not necessary. Once the listener is in what is considered the sweetspot based on virtual imaging of centrally derived images, then 10 uSec R-L delay of other signals will cause the other signals to be positioned laterally off center.
See previous explanation.
In excess of 660 C? Or, 1062 C? Or, did it exceed glass transition?
cmon scott...need pics..
Cheers, jn
T
No close mic'ing needed, the band is loud enough you can mic sections and get a decent blend.
I really don't care what analog's range is, I want the best! So 22 real bits with 8 x oversampling should meet requirements for the recording system not to be the limiting factor.
Flat is really not good on an SLM as there are a lot of infrasonic noise sources that are very very loud, some LF roll off is required.
Finally Apples are computers and Oranges are American football players, no comparison at all!
No close mic'ing needed, the band is loud enough you can mic sections and get a decent blend.
I really don't care what analog's range is, I want the best! So 22 real bits with 8 x oversampling should meet requirements for the recording system not to be the limiting factor.
Flat is really not good on an SLM as there are a lot of infrasonic noise sources that are very very loud, some LF roll off is required.
Finally Apples are computers and Oranges are American football players, no comparison at all!
Howard,
My guess would be that this is an example of the alias frequency of a midband signal being the same as the correct signal. So I would expect a 6 db rise in garbage at the midband.
There might be some spreading due to clock jitter.
ES
I can't position myself within 0.3mm. Yet my speakers image just fine, and moving imperceptibly does not shift the image.
SY
There was a fellow who taught that coaxial speakers were the solution to the path length difference issue and frequency notches that resulted at crossover.
So when I place my high frequency horn 24" away from the center of the midrange horn, what is the path length difference to a close listener at say 200 feet? (Hint the cancellation notch would occur at 56 khz if both drivers had similar outputs.)
As you indicated before there are lots of room reflections in the absolute timing issue to be wary off. But that Pythagoras cat did some really cool math way back even before John was born.
Your response is again, non sequitur.
I am speaking of a differential positioning between a central image with ITD and IID of zero, and an imaged voice shifted as a result of a 10 uSec ITD with respect to the image with none.
You are mistakenly attempting to introduce absolutes into a referenced discussion.
Cheers, jn
Hi,
So, ahead of the conductors rostrum...
Do not get me wrong, I would love a 22 Bit SAR or Flash ADC sampling at 384KHz for recordings and I would love to store the masters in this format.
I am pragmatic though, CD can be made very good (but rarely is) and I'd rather have the mainstream recording industry use 16/44 to it's limits than two or three audiophile labels issuing superb recordings in 22/384 of mediocre performances.
Not flat to DC, not sure where precisely the LF rolloff ended up, I normally aim for 8 Hz per stage, if I remember right it had around six rolloffs. So there was an infrasonic rolloff.
My definition of "Flat" is "without weighting" but "suitably bandwidth limited".
My point precisely.
Ciao T
So, ahead of the conductors rostrum...
Do not get me wrong, I would love a 22 Bit SAR or Flash ADC sampling at 384KHz for recordings and I would love to store the masters in this format.
I am pragmatic though, CD can be made very good (but rarely is) and I'd rather have the mainstream recording industry use 16/44 to it's limits than two or three audiophile labels issuing superb recordings in 22/384 of mediocre performances.
Not flat to DC, not sure where precisely the LF rolloff ended up, I normally aim for 8 Hz per stage, if I remember right it had around six rolloffs. So there was an infrasonic rolloff.
My definition of "Flat" is "without weighting" but "suitably bandwidth limited".
My point precisely.
Ciao T
- Status
- Not open for further replies.