Filters and Processors
You're right that the output inductors are a little trickier to source. This is one of the final details I'm sorting out before I order my boards.
TI recommends a shielded bobbin type inductor. Most 384KHz Equibit amps with BTL configured outputs require a 10uH inductor coming off of each half-bridge. This gives a total of 20uH for each channel, the same as for the Zetex parts you were using. I'm guessing they must have had a similar carrier frequency. Their 30W reference design uses a Taiyo Yuden LHFP13BB100M, which just happens to be the largest value available for this part.
Once you get up to a certain power level (say >50VA) there are enough variables with transformers and inductors that it isn't practical for manufacturers to offer off the shelf parts. Instead they have a selection of standard cores and do custom winding. Many of these places are set up to do low volume samples.
I'm also considering using output inductors wound on torroid cores.
The other part I'm finding slightly tricky (at least in small quantities) are the helically trimmed MELF0204 resistors required for the power supply snubbers. I'll do a more complete search for these once I settle on the values I require.
Yes, I have this book. It's excellent!
While it isn't specifically for audio, it's probably the best single DSP book I've found. Even though it's offered free on the website, I highly recommend buying the printed copy and supporting the author, Steven Smith.
Microchip's dsPICs should be really usefull for digitally generating test signals and tuning a system, but they don't have sufficient horsepower for much processing of the actual audio signal. I've got the development system for these, although I haven't started using it yet.
In case you didn't see it, there was a thread that included a discussion of audio processors:
Digital speaker phase correction?
I'm planning to start with TI's TAS3103 Digital Audio Processor, which is a preprogrammed DSP for audio. The program can't be changed on this part. The only thing that can be programmed are the filter and gain coefficients. This makes it much easier to get going with this part than with a general purpose DSP. You can start with all of the default values and just use it as a volume control. The other functions can then be experimented with one at a time when you're ready.
The only shortcoming that I see with the TAS3103 (and this is true of most other DSPs), is that it doesn't have the capability of doing massive custom FIRs. Once I'm ready to experiment with these, I plan to use an FPGA. A high powered DSP (such as the TMS320DA610, when it becomes generally available) would be another option. Other than the FIRs, the TAS3103 has the processing bandwidth and features to do everything else I anticipate wanting to do.
Regards,
Brian.
Rarkov said:
You're right that the output inductors are a little trickier to source. This is one of the final details I'm sorting out before I order my boards.
TI recommends a shielded bobbin type inductor. Most 384KHz Equibit amps with BTL configured outputs require a 10uH inductor coming off of each half-bridge. This gives a total of 20uH for each channel, the same as for the Zetex parts you were using. I'm guessing they must have had a similar carrier frequency. Their 30W reference design uses a Taiyo Yuden LHFP13BB100M, which just happens to be the largest value available for this part.
Once you get up to a certain power level (say >50VA) there are enough variables with transformers and inductors that it isn't practical for manufacturers to offer off the shelf parts. Instead they have a selection of standard cores and do custom winding. Many of these places are set up to do low volume samples.
I'm also considering using output inductors wound on torroid cores.
The other part I'm finding slightly tricky (at least in small quantities) are the helically trimmed MELF0204 resistors required for the power supply snubbers. I'll do a more complete search for these once I settle on the values I require.
This is turning out to be quite interesting!Have you tried looking at the DSP Guide. It is a free online book of DSP. Very useful!
Yes, I have this book. It's excellent!
While it isn't specifically for audio, it's probably the best single DSP book I've found. Even though it's offered free on the website, I highly recommend buying the printed copy and supporting the author, Steven Smith.
Microchip's dsPICs should be really usefull for digitally generating test signals and tuning a system, but they don't have sufficient horsepower for much processing of the actual audio signal. I've got the development system for these, although I haven't started using it yet.
In case you didn't see it, there was a thread that included a discussion of audio processors:
Digital speaker phase correction?
I'm planning to start with TI's TAS3103 Digital Audio Processor, which is a preprogrammed DSP for audio. The program can't be changed on this part. The only thing that can be programmed are the filter and gain coefficients. This makes it much easier to get going with this part than with a general purpose DSP. You can start with all of the default values and just use it as a volume control. The other functions can then be experimented with one at a time when you're ready.
The only shortcoming that I see with the TAS3103 (and this is true of most other DSPs), is that it doesn't have the capability of doing massive custom FIRs. Once I'm ready to experiment with these, I plan to use an FPGA. A high powered DSP (such as the TMS320DA610, when it becomes generally available) would be another option. Other than the FIRs, the TAS3103 has the processing bandwidth and features to do everything else I anticipate wanting to do.
Regards,
Brian.
Thoughts on switching frequency
A number of opponents to digital amplifiers call out the switching frequency of current generation of class D amplifiers as a problem. Discussions about distortion caused by the output filter, difficulty in getting the output filter correct, its heat dissipation, high switching frequencies > 2 Mhz needed to be truly load invariant are some of the issues discussed. From Brian's previous explainations of the technologies involved (Equibit with its skewed PWM timings, Tripath with its spread spectrum & adaptive technologies), is it fair to assume that you cannot compare implementations based on switching frequency alone?
However, in general the higher the switching frequency, the easier it would be to design a high performance output filter. The TI chips @ 384Khz look to be at a disadvantage compared to some of the other designs. With most things in audio, its not the theory that counts, its the implementation which is why it would be great to evaluate the sonic differences between the different implementations. I'm looking forward to reading Brian's comments on his experements with the TI chips as they sound like they have great potential
Comments?
Regarding varying the power supply for volume, a thought that comes to mind to do this simply is to use a Varac which will save the complexity of a variable voltage switching power supply.
One other thing that requires careful consideration is the amount of digital hash created by dumping large currents at high frequencies, which can radiate lots of RF noise as well as pollute the mains supply. Careful layout & filter design is needed here.
Dean
A number of opponents to digital amplifiers call out the switching frequency of current generation of class D amplifiers as a problem. Discussions about distortion caused by the output filter, difficulty in getting the output filter correct, its heat dissipation, high switching frequencies > 2 Mhz needed to be truly load invariant are some of the issues discussed. From Brian's previous explainations of the technologies involved (Equibit with its skewed PWM timings, Tripath with its spread spectrum & adaptive technologies), is it fair to assume that you cannot compare implementations based on switching frequency alone?
However, in general the higher the switching frequency, the easier it would be to design a high performance output filter. The TI chips @ 384Khz look to be at a disadvantage compared to some of the other designs. With most things in audio, its not the theory that counts, its the implementation which is why it would be great to evaluate the sonic differences between the different implementations. I'm looking forward to reading Brian's comments on his experements with the TI chips as they sound like they have great potential
Comments?
Regarding varying the power supply for volume, a thought that comes to mind to do this simply is to use a Varac which will save the complexity of a variable voltage switching power supply.
One other thing that requires careful consideration is the amount of digital hash created by dumping large currents at high frequencies, which can radiate lots of RF noise as well as pollute the mains supply. Careful layout & filter design is needed here.
Dean
Re: Thoughts on switching frequency
There are two advantages to a higher switching frequency:
Greater dynamic resolution and smaller low pass reconstruction filters.
By playing tricks with the timing, Equibit achieves a much higher resolution for a given PWM frequency than would standard fixed PWM.
In a BTL configuration, Equibit can use fully symetrical, balanced filters. Many Class D amps use unbalanced filters. The use of balanced filters greatly reduces distortion, and also makes filter design somewhat easier.
A lot of the problems Class D has arise from the feedback loop. The circuitry can only make one 'correction' per PWM pulse. This greatly limits the response bandwidth. There are also issues with the way the feedback circuitry interacts with the output filter. Higher switching frequencies help with these issues, but then cause other issues with power disipation, EMI, and other types of distortion. As I mentioned previously, an output device has pretty much fixed turn-on and turn-off times. Higher switching frequencies mean that the required underlap time will be a greater percentage of the total time. This increases distortion.
Tripath deals with the 'one feedback correction per pulse' limitation by using an adaptive / predictive filter. It learns from its past mistakes and doesn't strictly rely on instantanious feedback.
Equibit completely avoids these issues by running open loop.
*****
Equibit amps do have load sensitivity with regard to load impedance. Lower impedance speakers will cause the individual current spikes through the output FETs to be greater. One way to deal with this is to overrate the output devices to handle the worst-case load. To adjust an amp for an individual speaker, the main parameter is the supply voltage. The optimum voltage would be set so that the speaker achieves its maximum output at around 80% to 90% PWM duty cycle. (A lower voltage wouldn't allow for maximum output - but could be used for volume control. A higher voltage would simply increase the current spikes in the FETs.)
Placing the amps near the speaker is desirable to eliminate long speaker wire runs. It's better to physically distribute while the signal is still digital.
There are many good reasons to design an audio system with multiple Equibit amps - one per speaker driver.
1.) Digital crossovers with amplitude and phase correction for each driver.
2.) Optimized supply voltage for each driver.
3.) Elimination of passive crossovers.
Besides all of the advantages of digital crossovers, this allows the amp to have much better control of the driver. Equibit amps have an excellent damping factor. Any back-EMF from the driver gets directly shunted to the power supply. (A passive crossover reduces the coupling to the amp and can also allow back EMF from one driver into another.) Also, a passive crossover usually has a greater voltage drop than the output devices in the amp. Because an Equibit amp has such high efficiency to begin with, elimination of the passive crossover is one of the major opportunities to further increase efficiency.
4.) It's easier to build multiple lower power output sections than it is to build a single higher power section.
You're certainly right about that.
In my other life, I work on industrial motor drive design. One of the little jokes we have is that it is a good idea never to get a reputation for being good at dealing with EMI issues. If you get identified as being some type of EMI guru, you will be forever sentenced to chasing the little demons.
I'm relying heavily on TI's reference design for the board layout of the output section. They clearly have paid close attention to power routing. There are little RC snubber circuits sprinkled all over the place. The component values for many of these get fine tuned after the board is built. The main power input to the TAS5110 has a reactive LC filter using controlled board traces for the inductors.
I don't think I would have even considered attempting to build one of these if I didn't have the reference design as an example.
Regards,
Brian.
I would think so.deandob said:
There are two advantages to a higher switching frequency:
Greater dynamic resolution and smaller low pass reconstruction filters.
By playing tricks with the timing, Equibit achieves a much higher resolution for a given PWM frequency than would standard fixed PWM.
In a BTL configuration, Equibit can use fully symetrical, balanced filters. Many Class D amps use unbalanced filters. The use of balanced filters greatly reduces distortion, and also makes filter design somewhat easier.
A lot of the problems Class D has arise from the feedback loop. The circuitry can only make one 'correction' per PWM pulse. This greatly limits the response bandwidth. There are also issues with the way the feedback circuitry interacts with the output filter. Higher switching frequencies help with these issues, but then cause other issues with power disipation, EMI, and other types of distortion. As I mentioned previously, an output device has pretty much fixed turn-on and turn-off times. Higher switching frequencies mean that the required underlap time will be a greater percentage of the total time. This increases distortion.
Tripath deals with the 'one feedback correction per pulse' limitation by using an adaptive / predictive filter. It learns from its past mistakes and doesn't strictly rely on instantanious feedback.
Equibit completely avoids these issues by running open loop.
*****
Equibit amps do have load sensitivity with regard to load impedance. Lower impedance speakers will cause the individual current spikes through the output FETs to be greater. One way to deal with this is to overrate the output devices to handle the worst-case load. To adjust an amp for an individual speaker, the main parameter is the supply voltage. The optimum voltage would be set so that the speaker achieves its maximum output at around 80% to 90% PWM duty cycle. (A lower voltage wouldn't allow for maximum output - but could be used for volume control. A higher voltage would simply increase the current spikes in the FETs.)
Placing the amps near the speaker is desirable to eliminate long speaker wire runs. It's better to physically distribute while the signal is still digital.
There are many good reasons to design an audio system with multiple Equibit amps - one per speaker driver.
1.) Digital crossovers with amplitude and phase correction for each driver.
2.) Optimized supply voltage for each driver.
3.) Elimination of passive crossovers.
Besides all of the advantages of digital crossovers, this allows the amp to have much better control of the driver. Equibit amps have an excellent damping factor. Any back-EMF from the driver gets directly shunted to the power supply. (A passive crossover reduces the coupling to the amp and can also allow back EMF from one driver into another.) Also, a passive crossover usually has a greater voltage drop than the output devices in the amp. Because an Equibit amp has such high efficiency to begin with, elimination of the passive crossover is one of the major opportunities to further increase efficiency.
4.) It's easier to build multiple lower power output sections than it is to build a single higher power section.
You're certainly right about that.
In my other life, I work on industrial motor drive design. One of the little jokes we have is that it is a good idea never to get a reputation for being good at dealing with EMI issues. If you get identified as being some type of EMI guru, you will be forever sentenced to chasing the little demons.
I'm relying heavily on TI's reference design for the board layout of the output section. They clearly have paid close attention to power routing. There are little RC snubber circuits sprinkled all over the place. The component values for many of these get fine tuned after the board is built. The main power input to the TAS5110 has a reactive LC filter using controlled board traces for the inductors.
I don't think I would have even considered attempting to build one of these if I didn't have the reference design as an example.
Regards,
Brian.
Hi there,
I'm considering building one of these amps myself, with the TI devices and descrete output switching FETs. I haven't got that far into the design yet though, but it's beutiful to see your nowledge here!
A few questions for you guys (mostly Brian so far?) which has gotten a little bit further in designing:
1) Why not consider the higher spec parts TAS5015 and TAS5182? Cos they are not available/sampling yet? Layout should be similar to the parts on the EVM board.
2) The soldering of these parts will not be trivial with the 0.65mm spacing. Do you have professional equipment at hand or do you have other smart ways of dealing with it? Otherwise I can see this becoming a potential showstopper here....
3) Brian, as I understand it you are not buying the EVM, but designing you own PCB? Are you considering posting you PCB's files here? I could use a starting/branching point...
4) It seems like the TacT Millenium uses quite large Jensen air core foil inductors placed close to the speaker terminals. I think it could be a good choice if the distance from the switching stage is not a big issue. They seems bigger than some 10uH though?
(pics)
Jensen
5) What are you considering for high quality volume control? Are you all planning a digital attenuation (potentially decreasing the resolution) or is anybody have ideas for a H.Q. variable power supply?
6) My plan is a totally digital amp with the reciever chip CS8420, which can use an external precision oscillator for clock generation. Any comments on the need for an very stable clock, and its placement?
cheers all,
I'm considering building one of these amps myself, with the TI devices and descrete output switching FETs. I haven't got that far into the design yet though, but it's beutiful to see your nowledge here!
A few questions for you guys (mostly Brian so far?) which has gotten a little bit further in designing:
1) Why not consider the higher spec parts TAS5015 and TAS5182? Cos they are not available/sampling yet? Layout should be similar to the parts on the EVM board.
2) The soldering of these parts will not be trivial with the 0.65mm spacing. Do you have professional equipment at hand or do you have other smart ways of dealing with it? Otherwise I can see this becoming a potential showstopper here....
3) Brian, as I understand it you are not buying the EVM, but designing you own PCB? Are you considering posting you PCB's files here? I could use a starting/branching point...
4) It seems like the TacT Millenium uses quite large Jensen air core foil inductors placed close to the speaker terminals. I think it could be a good choice if the distance from the switching stage is not a big issue. They seems bigger than some 10uH though?
(pics)
Jensen
5) What are you considering for high quality volume control? Are you all planning a digital attenuation (potentially decreasing the resolution) or is anybody have ideas for a H.Q. variable power supply?
6) My plan is a totally digital amp with the reciever chip CS8420, which can use an external precision oscillator for clock generation. Any comments on the need for an very stable clock, and its placement?
cheers all,
Strummer said:
The TAS5015 is a substantially more complicated part to use, especially in terms of the gate drive, protection circuitry, and power supply requirements. The layout for this will be very different than the EVM. Also, when I first started my design, it was only available to certain Equibit licensees. (It appears to be available for general consumption now.)
The TAS5182 isn't available yet. I plan on using it for future designs.
I use a Metcal SP200 soldering iron with an old used Leica binocular scope.
I use ChipQuick for removal of high lead density parts.
I also have a small oven that is useful for preheating or reflowing boards.
Technique is the most important thing when working with fine pitch surface mount components, much more than specialized equipment. In a pinch, I've gotten excellent results from a cheap 15W soldering iron with the tip filed down to a small blade.
For me, getting the right dose of coffee is critical. Not enough and I can't concentrate. Too much and I get jittery.
The bottom line is to get lots of practice. Learn to be quick and don't be willing to settle for anything less than an ideal joint: fully wetted with no peaking.
A good stereo microscope is probably the best place to spend limited money. I bought mine used for a couple hundred bucks. I can do fine pitch parts without one, but usually get a headache. Also, I'll mention that I find magnifying lenses to be worse than nothing at all. I have yet to find one with a steady enough holder.
Anybody that wants to play with new technology components has to come to terms with working on fine-pitched devices. There is virtually nothing new being introduced with old-style packages.
My next challenge is figuring out how to deal with BGA and chip-scale packages at home.
I'll have to answer that with a definite Maybe...
First I want to get something that works to my satisfaction.
Frankly, I suspect that someone wanting to do anything other than an exact copy of one of my boards would do just as well to start with TI's reference layout.
Thanks for pointing this out.
I looked at Jensen's website. I think their inductors would be useful for an Equibit output with discrete transistors. I'm afraid they have too much series resistance for the integrated TAS5110.
Their four pole electrolytics also look promising for later higher powered designs.
I discussed this somewhat on earlier posts in this thread. I currently am doubting that digital volume control will cause any meaningful decrease in resolution. To experiment with this, I'm still designing my first board so that I can vary the supply level.
Preference should be given to locating the clock near the Equibit modulator (i.e. TAS5012). Think of it as a DAC. I would still try to get the routing as direct and close to the CS8420 as possible.
Regards,
Brian.
Brian,
I have one more question to you, because it seems that you are the smartest guy on this thread.
One more thing about TAS5015 is not quite clear to me.
According to datasheet TAS5015 requires an external PLL, but when operating in master mode does this realy need to be a PLL or can it be a crystal oscillator? My understanding of this is that when operating in slave mode the high frequency clock signal on pin 6 must be phase locked to LRCLK, but when operating in master mode there shouldn't be any specific phase relationship between HFCLK and any other clock signal, because the LRCLK, SCLK and MCLK are outputs derived from HFCLK.
Is this correct?
I'm interested in master mode operation because I would like to use TAS5015 with AD1896 asynchronous sample rate converter.
Best regards,
Dejan
I have one more question to you, because it seems that you are the smartest guy on this thread.
One more thing about TAS5015 is not quite clear to me.
According to datasheet TAS5015 requires an external PLL, but when operating in master mode does this realy need to be a PLL or can it be a crystal oscillator? My understanding of this is that when operating in slave mode the high frequency clock signal on pin 6 must be phase locked to LRCLK, but when operating in master mode there shouldn't be any specific phase relationship between HFCLK and any other clock signal, because the LRCLK, SCLK and MCLK are outputs derived from HFCLK.
Is this correct?
I'm interested in master mode operation because I would like to use TAS5015 with AD1896 asynchronous sample rate converter.
Best regards,
Dejan
I'd like to throw in my 2 pence about the tradeoffs regarding the switching frequency of class-d amps (either digital like Equibit or the classic analog PWM).
It is indeed true that for a given filter the carrier supression is better with a higher switching frequency.
OTOH magnetic core materials that can handle high frequencies with low losses are less widely available than those for lower switching frequencies. I was once thinking about air-cored output filters (the Tact actually uses such) but then you have additionall EMC issues. A compromise would be the use of air-gaps as we did within a PWM amplifier 10 years ago.
The MOSFET and driver switching losses also increase with frequency.
A given absolute timing error means a larger relative timing error for a higher switching frequency and therefore increased distortion. OTOH a higher switching frequency allows for a higher unity-gain point and therefore more NFB.
Once it was thought that the complete (whatever that is) removal of the carrier is important but noone cares that much about that nowadays anymore, to which I can agree as well (our amp had a carrier suppression of > 80 dB).
Regards
Charles
It is indeed true that for a given filter the carrier supression is better with a higher switching frequency.
OTOH magnetic core materials that can handle high frequencies with low losses are less widely available than those for lower switching frequencies. I was once thinking about air-cored output filters (the Tact actually uses such) but then you have additionall EMC issues. A compromise would be the use of air-gaps as we did within a PWM amplifier 10 years ago.
The MOSFET and driver switching losses also increase with frequency.
A given absolute timing error means a larger relative timing error for a higher switching frequency and therefore increased distortion. OTOH a higher switching frequency allows for a higher unity-gain point and therefore more NFB.
Once it was thought that the complete (whatever that is) removal of the carrier is important but noone cares that much about that nowadays anymore, to which I can agree as well (our amp had a carrier suppression of > 80 dB).
Regards
Charles
Rarkov said:
A clever way to get help from one person at the expense of anyone else participating!
Gaz
I may have written the most in this thread so far, and I'll have to admit that I've probably gone off the deep end with the amount of time that I've spent on this project the last half year, but remember I'm still working on my first attempt. It'll be a major milestone for me just to get the thing functional with some sort of sound coming out. Only then can I start playing around with options to hopefully furthur improve the sound.
I really appreciate and want to hear from others with switching amplifier experience, or that are experimenting with or thinking about digital amps.
Before I recently joined this forum, I had assumed that there would be many people who had already used TI's Equibit chips. Digikey has been selling them for over a year now! (Which I've found to be one of the best indicators that a new component or technology has become accepted.) It's really surprised me that I haven't been able to find a single person (even outside this forum) who has already built a home-brew version of one of these.
It's certainly not because a digital amp is more ambitious than other projects people are doing here. I see other people doing far more advanced things taking even greater efforts.
Perhaps it's because this is still kind of new. I'm very optimistic about the potential that digital amps offer for achieving new levels of sonic performance. I would think many people would want to try it out.
So please, speak up!
Better yet, start playing with this stuff.
Regards,
Brian.
Rookie said:
This is very correct.
I'd only realized this point a couple of weeks ago.
Using the AD1896 ASRC allows the digital amp to be a clock master, so as you point out it's possible to use an oscillator instead of an external PLL with the TAS5015. The TAS5015 can then generate all of the other audio clocks in Master mode.
I noticed something else as well:
HFCLK (p6 on TAS5015) is NC on TAS5012.
All of the pins associated with the internal PLL (p3, p4, p45, p46) on the TAS5012 are NCs on the TAS5015.
This means I can lay my first board out to use either the TAS5012 or the TAS5015! I can place oscillator footprints right next to p6 (for the TAS5015) and p1 (MCLKIN on the TAS5012).
Normally, for a 96KHz or 192KHz sampling rate:
The TAS5012 requires a 12.288 MHz oscillator.
The TAS5015 requires a 98.304 MHz oscillator.
The 12.288 MHz is a reasonably standard value. I'm going to use a CTS CB3LV-3C-12.2880-T ($5.08 from Digi-Key). This part is spec'd to have <1pS jitter.
98.304 MHz isn't a standard value. This leaves three options:
1.) Use a custom part (not now for me, at least).
2.) Use a programmable oscillator (I don't know of any with as good jittter specs - this defeats some of the purpose of what we're trying to achieve).
3.) Operate the system at a non-standard frequency (the AD1896 makes this possible).
Option 3 seems like the clear choice to me, but there's a problem:
The closest standard oscillator frequency to 98.304 MHz is 100MHz. This would produce a system fs of 97.656 KHz or 195.313KHz, with an SCLK frequency of 12.5MHz. The TAS5015's data sheet specifies 12.288MHz as the maximum SCLK. I'm guessing that this is listed because it's a standard value and that the absolute limit of the chip is somewhat higher. As long as the other setup and hold time requirements are met by the other components in the system, my guess is that it shouldn't be a problem. I guess it all depends on whether or not you're ready for membership in Ov&rcl@ckers Anonymous.
The next standard frequency down in ultra-low jitter oscillators is 90MHz. I think this might be a significant audio quality hit to fs, but maybe not.
So that's where it's at with me. This is one of the things I plan to play around with, take measurements, and listen to.
Regards,
Brian.
phase_accurate said:
Thanks, it's good to hear from someone who's worked with these things.
Any design involves making compromises, but it strikes me that with switching amplifiers it's even more of a balancing act.
One can't help but notice the enclosure that TacT uses. Unlike a lot of high end amps, I'm sure the TacT's is more than just eye-candy.
TI claims their reference design (with shielded bobbins and the circuit board mounted over an aluminum) passes CE emisions requirements outside of an enclosure. Pretty impressive for a fairly high power, high frequency switcher.
This does seem reasonable, as long as any remaining power from the carrier isn't significantly cutting into the amp's efficiency or the driver's power handling.
So are you presently doing anything in this area?
Regards,
Brian.
About oscillators
You can request samples of crystal oscillators from Fox Electronics and MF Electronics at any specified frequency.
www.foxonline.com
www.oscillators.com
You can request samples of crystal oscillators from Fox Electronics and MF Electronics at any specified frequency.
www.foxonline.com
www.oscillators.com
Re: About oscillators
Thanks for the links!
Fox doesn't list any jitter specs for their oscillators, so I'm assuming they don't have anything in the ultra-low jitter catagory.
MF Electronics has several products in the 100MHz range with jitter in the 5-6pS range, which is as good as I've seen from any company at this frequency. I just might give their custom sample service a try.
******
FYI:
With a regular DAC, jitter will affect the timing of the output. Increased jitter causes a type of IM distortion.
With a pwm-based digital amp, jitter affects output timing in the same way as a regular DAC, but it will also reduce dynamic range. (PWM output level is controled by variations in output timing.)
The main performance spec differences between the TAS5012 and TAS5015 are dynamic range and THD+N:
TAS5012:
dynamic range: 102-db
THD+N: 0.06%
TAS5015:
dynamic range: 112-db
THD+N: 0.01%
Using a variable-voltage power supply has the potential of increasing the dynamic range of either by an additional 30-db!
I believe the only principle difference that gives the TAS5015 better performance is lower jitter levels. It relies on the use of an external PLL (or oscillator) to generate its internal clock. The TAS5012 has a built-in PLL.
Wild speculation: It wouldn't surprise me at all if they used the same silicon. They might just be hooking up the bondout wires differently.
Just as with regular DACs, an Equibit stage typically would have to be sync'd to the incoming clock of the data source, hence the need for PLLs.
As we discussed earlier, the AD1896 ASRC allows us to have a local clock at the converter, which if done right will have less jitter than a clock recovered by a PLL. (A VXCO-based PLL might be an exception to this. I'm surprised I haven't seen anybody try this, even with a regular line-level DAC.)
(Another way to achieve jitter reduction is to have a high-quality master clock at the converter, and transmit it to the data source. The data source then sends the data to the converter along with a reflected clock. This data can then be sent through a FIFO to be reclocked by the original master clock. Since the reflected clock is sync'd with the master, there's no concern about a buffer underrun or overflow.)
Anyway, the specs in the TAS5012 and TAS5015 data are somewhat nebulus. Many data sheets list claims for best-case synarios, especially wrt front page claims. However, I'm speculating that in this case the claims are assuming a recovered clock. The story *might* change with stable master oscillators nearby.
So one of my questions comes down to:
Will it be better to use the TAS5012 with a 12.288MHz oscillator and <1pS jitter PLL'd times 8 (I'll use a local low-noise power supply for the PLL), or to use the TAS5015 with a 98.304MHz oscillator with 5pS jitter?
Enquiring minds want to know!
Regards,
Brian.
Rookie said:
Thanks for the links!
Fox doesn't list any jitter specs for their oscillators, so I'm assuming they don't have anything in the ultra-low jitter catagory.
MF Electronics has several products in the 100MHz range with jitter in the 5-6pS range, which is as good as I've seen from any company at this frequency. I just might give their custom sample service a try.
******
FYI:
With a regular DAC, jitter will affect the timing of the output. Increased jitter causes a type of IM distortion.
With a pwm-based digital amp, jitter affects output timing in the same way as a regular DAC, but it will also reduce dynamic range. (PWM output level is controled by variations in output timing.)
The main performance spec differences between the TAS5012 and TAS5015 are dynamic range and THD+N:
TAS5012:
dynamic range: 102-db
THD+N: 0.06%
TAS5015:
dynamic range: 112-db
THD+N: 0.01%
Using a variable-voltage power supply has the potential of increasing the dynamic range of either by an additional 30-db!
I believe the only principle difference that gives the TAS5015 better performance is lower jitter levels. It relies on the use of an external PLL (or oscillator) to generate its internal clock. The TAS5012 has a built-in PLL.
Wild speculation: It wouldn't surprise me at all if they used the same silicon. They might just be hooking up the bondout wires differently.
Just as with regular DACs, an Equibit stage typically would have to be sync'd to the incoming clock of the data source, hence the need for PLLs.
As we discussed earlier, the AD1896 ASRC allows us to have a local clock at the converter, which if done right will have less jitter than a clock recovered by a PLL. (A VXCO-based PLL might be an exception to this. I'm surprised I haven't seen anybody try this, even with a regular line-level DAC.)
(Another way to achieve jitter reduction is to have a high-quality master clock at the converter, and transmit it to the data source. The data source then sends the data to the converter along with a reflected clock. This data can then be sent through a FIFO to be reclocked by the original master clock. Since the reflected clock is sync'd with the master, there's no concern about a buffer underrun or overflow.)
Anyway, the specs in the TAS5012 and TAS5015 data are somewhat nebulus. Many data sheets list claims for best-case synarios, especially wrt front page claims. However, I'm speculating that in this case the claims are assuming a recovered clock. The story *might* change with stable master oscillators nearby.
So one of my questions comes down to:
Will it be better to use the TAS5012 with a 12.288MHz oscillator and <1pS jitter PLL'd times 8 (I'll use a local low-noise power supply for the PLL), or to use the TAS5015 with a 98.304MHz oscillator with 5pS jitter?
Enquiring minds want to know!
Regards,
Brian.
Hi Brian
You asked whether I am doing anything in the direction of a digital amp at the moment.
I actually developed a PWM amp for my thesis about 10 years ago. Back then there was almost no info available on that subject. But we got one up and running whithin the desired specs (there would have been room for improvement though). But I was already interested in switching amplifiers before that and that was the reason why I chose this subject out of a few dozens, when they were presented.
I am still interested in switching amplifiers, though not so much in digital ones (like the equibit) but classic PWM ones, but one should never say never .......
The main reason why I am interested in class-d amplifiers is their efficiency and not the "digital" appeal. I have indeed an application where I could use some efficient poweramps. Currently I am doing some studies regarding delta-sigma amplifiers (class-T and the sharp ones belong to this group).
I knew that these TI chipsets exist but didn't know that they are available to everyone now, because TI first intended to sell them only to large companies and only if they pay huge licence fees.
I had the opportunity to listen to the TacT millennium (and also the newer TacT amplifiers including TacT's room correction system) and I must admit that it is definitely an outstanding amp. The amp's design is indeed an excellent showcase from both the technical AND aesthetical point of view. One thing I already mentioned somewhere else is its low end punch (which is theoretically inherent to any well made switching amp ). I assume the high sound quality doesn't come from the working principle alone. I have the distinct feeling that there is much more room for doing things wrong compared with ordinary linear amplifiers. A linear amp where the same care has been put into will also be excellent ........
If there are general questions regarding class-d (i.e. not the equibit chipsets themselves) or if somebody is interested in my "old one" I could of course assist.
Regards
Charles
You asked whether I am doing anything in the direction of a digital amp at the moment.
I actually developed a PWM amp for my thesis about 10 years ago. Back then there was almost no info available on that subject. But we got one up and running whithin the desired specs (there would have been room for improvement though). But I was already interested in switching amplifiers before that and that was the reason why I chose this subject out of a few dozens, when they were presented.
I am still interested in switching amplifiers, though not so much in digital ones (like the equibit) but classic PWM ones, but one should never say never .......
The main reason why I am interested in class-d amplifiers is their efficiency and not the "digital" appeal. I have indeed an application where I could use some efficient poweramps. Currently I am doing some studies regarding delta-sigma amplifiers (class-T and the sharp ones belong to this group).
I knew that these TI chipsets exist but didn't know that they are available to everyone now, because TI first intended to sell them only to large companies and only if they pay huge licence fees.
I had the opportunity to listen to the TacT millennium (and also the newer TacT amplifiers including TacT's room correction system) and I must admit that it is definitely an outstanding amp. The amp's design is indeed an excellent showcase from both the technical AND aesthetical point of view. One thing I already mentioned somewhere else is its low end punch (which is theoretically inherent to any well made switching amp ). I assume the high sound quality doesn't come from the working principle alone. I have the distinct feeling that there is much more room for doing things wrong compared with ordinary linear amplifiers. A linear amp where the same care has been put into will also be excellent ........
If there are general questions regarding class-d (i.e. not the equibit chipsets themselves) or if somebody is interested in my "old one" I could of course assist.
Regards
Charles
The TAS5012 costs 9$ at Digikey and the TAS5015 costs 32$. Assuming this I think that there must be some bigger difference between them, not only the internal or external PLL. And who knows how Equibit actually works?It is a big secret. Maybe the TAS5015 use better Equibit modulator than the TAS5012.
I would go for TAS5015 with a little higher jitter clock instead of
TAS5012 with ultra low jitter master clock. The internal PLL of TAS5012 still has to derive the high frequency clock from MCLK and its intrinsic jitter is higher for sure. Maybe it's on a level of DIR1703 which has 75 psec of jitter. I'm not sure for this.
This is from Hifi Choice review of TacT Millennium:
"The outboard DAC analogy extends to the Millennium's dependence on a digital source - typically a CD transport - and, like other two-box combinations, it has to deal with the jitter that's aggravated in between. The black trace on my jitter plot (Figure 2) not only shows a high level of power supply-related jitter (4) - to 900psec - but also a broad hump (5) that's never good news for sound quality."
900 psec of jitter! And it still sounds great!
Regards
I would go for TAS5015 with a little higher jitter clock instead of
TAS5012 with ultra low jitter master clock. The internal PLL of TAS5012 still has to derive the high frequency clock from MCLK and its intrinsic jitter is higher for sure. Maybe it's on a level of DIR1703 which has 75 psec of jitter. I'm not sure for this.
This is from Hifi Choice review of TacT Millennium:
"The outboard DAC analogy extends to the Millennium's dependence on a digital source - typically a CD transport - and, like other two-box combinations, it has to deal with the jitter that's aggravated in between. The black trace on my jitter plot (Figure 2) not only shows a high level of power supply-related jitter (4) - to 900psec - but also a broad hump (5) that's never good news for sound quality."
900 psec of jitter! And it still sounds great!
Regards
Rookie said:
The cost may or may not be related to the cost of actually producing the part. It might just be a marketing decision. If there is a difference between them besides the PLL, it's probably that the TAS5015 has a larger FIR for interpolation.
Once I actually get something working, I think that it should be easy to tell if there's any actual processing difference by scoping the signals and feeding in some test code pulses.
Initially I don't want to sacrifice any parts. If I accidently damage a chip I might decap them to see if the silicon's different.
The general implementation of Equibit isn't a secret.
Like other oversampling DACs, the incoming data is interpolated to a higher sampling rate.
This high frequency PCM data is convolved into PWM by a single FIR filter. This FIR is look-up table based. The relative timing of the upper and lower transistors in the H-bridge is independantly controlled so that finer resolution can be achieved than could be obtained by simple integer multiples of the (384KHz) carrier. Non-linearity of the PWM switching output is also considered in the development of the filter coeficients.
The only thing that is really proprietary is the specific number and values of the filter coeficients, and the process that was used to create them.
There's lots of information out there on the Web if you do a search. The old Tocatta website had lots of good information, but unfortunately this disappeared after TI bought them. (I'm not going to complain too much since TI is willing to sell me their chips!)
The quality of a PLL is primarily affected by the quality of its power supply, and by the original stability of the incoming clock that the PLL is trying to recover. The DIR1703's SpACT obtains a typical 75pS jitter output, but that's starting with a highly jittered S/PDIF signal. By starting with a <1pS jitter clock, the TAS5012's x 8 PLL should output jitter somewhere between 1pS and 8pS. This is highly dependant on the overall implementation (of course the jitter could be much worse). The only way to know is to try it. Even then, I'll have trouble measuring anything less than 20pS jitter.
Again, I'm planning on trying both. I'm not nearly as concerned about clock jitter in my first digital amp board as other things. (Power supply impedance, snubber circuits, the output filter, and switching dead-time are my biggest worries.)
I haven't seen this article, so I can't tell the whole context of this quote. However, I'm very sure the Millenium doesn't have 900pS of jitter controlling its Equibit modulator! TacT has clearly gone to great lengths in their clocking schemes.
TacT uses a reflected clocking scheme like I described previously:
If used with a TacT CD source, the Millenium provides the master clock and sends it back to the CD player. The CD sends the data to the Millenium along with a reflected clock. I'm guessing that the reviewer may have been looking at the reflected clock (which probably does have high jitter). This doesn't matter because the data is reclocked by the original master.
The S/PDIF and AES inputs use a recovered clock scheme.
Unless this reviewer was probing the internal clock signals right at the Equibit modulator, I don't think his comments would be accurate. Does the article contain any specifics about where he was examining the clock?
Regards,
Brian.
I haven't heard the TacT myself yet. I think I'd have to travel out of state. I want to sometime, though.
I'll be the first to admit that part of why I'm doing this project is to do something different. I'm learning tons of new stuff that will help me even if the sound quality of the amp isn't as good as I'm hoping.
Having said that, my expectations are very high for this project. Presuming that I can get all aspects of the amp to work well as a system, I think that it has the potential to exceed the sonic quality of a more traditional power supply, class A amp, and DAC for a given amount of money. Time will tell...
In the meantime, I just want to get something that works. I'm trying as much as I can to anticipate all of the issues affecting quality, and I'm trying to design the board for as much experimental flexibility as possible. I'll have a much better handle on things when I can start probing around active circuitry and listening to it. I thought really hard about spending the $499 on TI's eval board. I decided that even though it would take me much longer to start hearing it, that I would be better off starting with my own board. I'll have better features for what I want to experiment with, and it'll make for a much better head start on the second board.
Brian.
Re: Here is the link to the review:
Thanks for the link!
I couldn't see any of the plots being referenced, but that's OK.
The 900pS jitter mentioned was what was coming out of the NAD CD player. This clock is then recovered by a PLL inside the Millennium. They then go on to mention a reflected clock upgrade available for the NAD to work off of the TacT's clock.
I found this particularly interesting:
"But the final 11dB represents digital gain and can, depending on the peak level represented by the incoming digital data, cause the amp to crash into high distortion. Believe it or not, TACT was obliged to introduce this extra gain because various dealers were unsettled by the fact that the volume control could be advanced to full without creating distorted sound. "
One of the things that I've been wondering about is where to set the power supply voltage. Do I do it so that digital 0db is above the point of distortion, just at maximum output of the amp, or if the amp can overpower a driver (tweeters in particular) at the limit of the driver.
How do I decide the limit of a driver:
Rated power?
Audible breakup?
Some type of measured distortion threashold?
The analog part of my mind
still wants to have some sort of 'headroom' to be able to push things past the limits. Sort of like setting the level so a VU meter bounces into the red.
The digital part of my mind
tells me to set things for better dynamic range and no distortion.
Brian.
Rookie said:
Thanks for the link!
I couldn't see any of the plots being referenced, but that's OK.
The 900pS jitter mentioned was what was coming out of the NAD CD player. This clock is then recovered by a PLL inside the Millennium. They then go on to mention a reflected clock upgrade available for the NAD to work off of the TacT's clock.
I found this particularly interesting:
"But the final 11dB represents digital gain and can, depending on the peak level represented by the incoming digital data, cause the amp to crash into high distortion. Believe it or not, TACT was obliged to introduce this extra gain because various dealers were unsettled by the fact that the volume control could be advanced to full without creating distorted sound. "
One of the things that I've been wondering about is where to set the power supply voltage. Do I do it so that digital 0db is above the point of distortion, just at maximum output of the amp, or if the amp can overpower a driver (tweeters in particular) at the limit of the driver.
How do I decide the limit of a driver:
Rated power?
Audible breakup?
Some type of measured distortion threashold?
The analog part of my mind
still wants to have some sort of 'headroom' to be able to push things past the limits. Sort of like setting the level so a VU meter bounces into the red.The digital part of my mind
tells me to set things for better dynamic range and no distortion.Brian.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.