zawada@EN.ECN.PURDUE.EDU (Paul J Zawada) (11/21/89)
A day or two ago, I heard another one bit D-to-A algorithm from a strictly unreliable source. (A musician friend of mine.) Since 1 bit D-to-A was a recent topic, I thought I'd share this one with with the net. He claims this new revolutionary algorithm uses one bit to tell whether the output voltage should be incremented or decremented from its current state. i.e. Voltages are not quantisized by binary words. If the bit is a zero the exsisting voltage will be say, decremented by a given value. If the bit is a one, the voltage will be incremented by the same value. (I don't know if actually 0=decrement and 1=increment, but you get the idea.) Now I'm only an undergrad in EE and I haven't studied dsp much, but I cannot see how to actually implement this. I see a few problems. 1. What do you do if the voltage stays the same? Would you just go ahead and decrement or increment and hope that it doesn't screw you up in the longrun? 2. How fast would you take these one bit samples? I don't suppose the Nyquist sampling theorem would still hold since you are no longer just dealing with a voltage level. Take, for example a high frequency signal with a large amplitude. This causes a large jump in your output voltage (jump >> increment value). You will need a number of samples to reflect such a jump. Conversely, if you are sampling at a very high rate, you will run into problem #1 with low frequency signals...You will have points at which the voltage doesn't change. I can see an advantage however. This algorithm, coupled with a very small increment/decrement voltage, can allow for very high resolution which is not available with a standard 16 or even 32 bit D-to-A conversion. (But then again, you need a high sampling rate for high frequency signals, which can lead to problems with the amount of available memory..) Would anybody else like to comment? pjz... ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Paul J Zawada | zawada@ee.ecn.purdue.edu "E-site" Student Consultant | ...!pur-ee!ei.ecn.purdue.edu!zawada Purdue University | Engineering Computer Network | GO BOILERS!!!
jbm@eos.UUCP (Jeffrey Mulligan) (11/21/89)
zawada@EN.ECN.PURDUE.EDU (Paul J Zawada) writes: >A day or two ago, I heard another one bit D-to-A algorithm from a >strictly unreliable source. (A musician friend of mine.) Since >1 bit D-to-A was a recent topic, I thought I'd share this one with >with the net. He claims this new revolutionary algorithm uses one >bit to tell whether the output voltage should be incremented or >decremented from its current state. Sounds like a delta modulator (not new). >Now I'm only an undergrad in EE and I haven't studied dsp much, but >I cannot see how to actually implement this. I see a few problems. Input voltage is tracked on a capacitor. Capacitor voltage is compared to input. Comparator output dumps fixed current on/off the capacitor. -- Jeff Mulligan (jbm@aurora.arc.nasa.gov) NASA/Ames Research Ctr., Mail Stop 239-3, Moffet Field CA, 94035 (415) 694-3745
aurie@rhea.trl.oz.au (Alistair Urie - Radio and Satellite Networks) (11/21/89)
In article <8911201702.AA03844@en.ecn.purdue.edu> zawada@EN.ECN.PURDUE.EDU (Paul J Zawada) writes: > >A day or two ago, I heard another one bit D-to-A algorithm from a >strictly unreliable source. (A musician friend of mine.) Since >1 bit D-to-A was a recent topic, I thought I'd share this one with >with the net. He claims this new revolutionary algorithm uses one >bit to tell whether the output voltage should be incremented or >decremented from its current state. i.e. Voltages are not quantisized But this is good old delta modulation, one of the very first digital modulation systems invented. It works quite well provided you choose a bit rate that is high enough to handle the highest expected slope in the input signal. By the way I understand that NASA used it back in the Apollo days as the last ditch communications system - the last few minutes before complete comms blackout during re-entry. Alistair URIE Radio and Satellite Networks Phone: +61 3 541 6370 Telecom Research Laboratories Fax: +61 3 543 3339 770 Blackburn Rd. Clayton Vic Internet: aurie@rhea.trl.oz.au AUSTRALIA
sheppard@hpnmdla.HP.COM (Roger Sheppard) (11/22/89)
Yes this sounds like delta modulation but that is not new. The only recent anouncement that concerns a 1 bit D/A is from Phillips addressing the problem of costly and inaccurate 16/18 + D/A's on low level signals in CD players. They claim that by drastically oversampling at x256 they convert the parallel words to an 11.3MHz bit stream that represents a "pulse density modulation" that is similar to delta modulation in that it pumps up and down a filter. The rf components can be removed by gentle filtering because there is a wide frequency difference between the information and the data rate. Also adverse effects on the phase of inband signals are minimized. Since only two states are needed to go up or down a simple 1 bit D/A can be used and the filter designed with switched capacitor technology. Using VLSI, two chips can do the servo control, digital detection, correction, interpolation, oversampling, and D/A conversion for a fraction of the cost of todays players while maintaining the same S/N and better low level linearity.
cbm@well.UUCP (Chris Muir) (11/22/89)
Isn't a 1 bit D/A just a delta modulator? There have a few products
based on delta mod in the pro music field, most notably the TC2900 delay
line. It samples at 100Mhz (so they claim). It DOES sound wonderful.
--
_______________________________________________________________________________
Chris Muir | "There is no language in our lungs
{hplabs,pacbell,ucbvax,apple} | to tell the world just how we feel"
!well!cbm | - A. Partridgebryan@intvax.UUCP (Jon R Bryan) (11/23/89)
From article <7070002@hpnmdla.HP.COM>, by sheppard@hpnmdla.HP.COM (Roger Sheppard): > Yes this sounds like delta modulation but that is not new. The only recent > anouncement that concerns a 1 bit D/A is from Phillips addressing the Crystal Semiconductor in Austin, (512) 445 7222, manufactures several very interesting A/D converters, in case anyone is interested. One is a two-channel 16-bit Delta Sigma converter designed specifically for digital audio. Very nice parts, and I believe they go for around $100 quantity one. The "sample and holds," anti-aliasing and decimation filtersand voltage reference are all built in (28-pin package). -- Jon R. Bryan <=> bryan@intvax.UUCP Sandia National Laboratories Intelligent Machine Principles Division Albuquerque, New Mexico
legrady@ug.cs.dal.ca (Tom Legrady) (11/23/89)
This general family of conversions is known as Delta Modulation. As Paul Zawada says, you use a single bit to indicate whether the output should be incremented or decremented. The nice thing about the method is the simplicity of the conversions: On A->D, if the analog signal is greater (less than) the reference, then output a 1 (0) and increment (decrement) the value of the reference before the next sampling. On D->A, if the input is 1 (0), then increment (decrement) the counter and send the result to the DAC. If the signal has not changed at all, the result is least significant bit jitter as the reference signal goes just above, then just below the actual input. As you can see, this method eliminates successive aproximation loops and other complications. For the method to be successful, it is essential that the change in the signal be very small between samples. Normally this means that the sampling rate is much higher than with SA or flash conversion techniques. An ordinary converter at 44 KHz could change from one extreme value to the other from one sample to the next. This occurs only on the edge of Nyquists limit, and the presence of such changes makes it likely that signals over the limit could occur. Usually, signals change in an orderly manner. If that converter wasa 16 bit SA converter, the A->D was actually operating at about .7Mhz. A DM converter at .7Mhz could, in the same interval, only change by 16 samples. For a full-scale triangle wave input, the maximum frequency which can be followed is 10 Hz. Usually, the situation is much better, but this inefficiency is one of the reasons for the developement of Adaptive Delta Modulation techniqes. Essentially, these are two-bit converters: one bit says whether the value should be increased or decreased, and the other says whether the size of the increment should be increased or decreased. Thus while the basic DM degenerates to a ramp, the ADM degenerates to something closer to successive approximation. With these variations, the joints benefits of (relatively) simple one stage converters and decreased data storage requirements are obtained. If you want to process the music, as in a synthesizer or effects box, you're better off with parallel techniques. If your application calls for the storage of data in compressed form, and especially if your medium is serial to start with ADM type methods are ideal. That's why it was used in (I think it was the ..) DBX 700, an Industrial quality digital audio system which stored stereo audio of very high quality on a consumer Beta or VHS video tape, disguised as a video signal.
dean@image.soe.clarkson.edu (Dean Swan) (11/23/89)
From article <7070002@hpnmdla.HP.COM>, by sheppard@hpnmdla.HP.COM (Roger Sheppard): > anouncement that concerns a 1 bit D/A is from Phillips addressing the > problem of costly and inaccurate 16/18 + D/A's on low level signals > in CD players. They claim that by drastically oversampling at x256 they > convert the parallel words to an 11.3MHz bit stream that represents a > "pulse density modulation" that is similar to delta modulation in that > it pumps up and down a filter. The rf components can be removed by gentle > filtering because there is a wide frequency difference between the information > and the data rate. Also adverse effects on the phase of inband signals are > minimized. Since only two states are needed to go up or down a simple 1 bit > D/A can be used and the filter designed with switched capacitor technology. This technique is also not entirely new. As I posted a while back, when this thread started, The General Instruments SP0256 speech synthesizer chip has used this technique of D to A conversion for 5 or 6 years now. -Dean Swan dean@sun.soe.clarkson.edu
ingoldsb@ctycal.UUCP (Terry Ingoldsby) (12/05/89)
Which models and brands of CD players are using the single bit,
heavily oversampled technology? I saw in our local newspaper that
several vendors are pursuing this technique. True?
It would seem to me that the single bit technology would be a good
cost saving measure, since the tolerances in the D/A circuitry
would not have to be as precise as with an 18 bit D/A. Does this
mean I can expect single bit technology on the low end players soon?
--
Terry Ingoldsby ctycal!ingoldsb@calgary.UUCP
Land Information Systems or
The City of Calgary ...{alberta,ubc-cs,utai}!calgary!ctycal!ingoldsb