Bricasti Design CEO Brian Zolner opens up on the company’s flagship product, the M21 dual-mono DSD DAC.
As I am about to publish my findings on the Bricasti Design M21 DSD DAC, the complexity of the $21k DAC is such that Brian Zolner, CEO of the company, and I conducted many correspondences on the subject. The extent and detail of our discussion is such that I thought it would be a good read for all to enjoy before the review’s publishing. The following is a condensed version of our discussion.
Q: How does the M21 improve upon the M1?
Brian: The M21 has the same discreet output buffer found in the M1, but it goes a bit further. One way to think about the M21 is an extension or upgrade of the basic M1 design, by adding the Native DSD converter that we first designed for the M12 Source Controller, in a product that can be used as a DAC and not a pre/dac. Or one could also say it’s like the M12 without analog inputs. This would not fit in the M1 chassis, so we took the larger chassis from the M12, used superior analog power supplies, making it more powerful than the M1, and then added the ladder dac just because we wanted to see how it would perform and compare with our M1 sigma delta dac. So, there are 3 DACs in the M21.
Q: Please explain the significance of the three DACs.
Brian: For DSD conversion, we made a true DSD converter in the M21. It was first made for the M12 Source Controller, and part of the reason for making the M21 was to offer this Native DSD converter in a product that could be used as a DAC where as the M12 is a pre/dac product. This is a true DSD converter and is based on a very high speed switch followed by an analog filter; any mutes and level adjustments for transition noise removal is done in the analog domain. When a DSD signal is converted in the M21, there is no downsampling to PCM as in all other types of converters. DSD presents issues if not downsampled for use in most chips as it is not possible to perform any signal processing on it, so even a start and stop of the data stream presents issues for other DACs, like snaps, pops at full scale output. Most DACs convert the signal to PCM and they are then able to remove any issues like this in the digital domain. Bricasti Design does this all in the analog domain and this allows for seamless transitions between DSD and PCM streams. This is why the M21 has an analog attenuator, the main purpose of which is to elegantly manage the DSD issues noted above and integrate it with the data streams from the PCM DACs.
For PCM processing, there are 2 PCM converters in the M21.
One is a sigma delta type based on the ADI 1955, which is the same as used in the M1 and has been our standard DAC design in all our products starting with the M1. We only use this part for its final upsampling to 5-bit DSD or otherwise known as the sigma delta modulator, and we create the filters in our DSP and our own I-V or current-to-voltage conversion with high speed opamps. This part of the design is taken from the M1, and in here, there is no need to deal with transition noise issues as the DAC can manage things in addition to the DSP.
The third one is a Ladder DAC. This is based on a 20-bit ADI part, and again we create our anti-aliasing filters in our DSP and analog filtering, enabling the DAC to manage any transition issues in DSP or with the analog stage of the M21.
An important part of the product is the way we do our clocking and clock recovery: whatever part or design for the modulation one uses, a key factor is the clocking. We use our DDS (direct digital synthesis) technique to reclock the incoming data stream. We track the incoming Mclock and buffer and resync against our DDS clock. Yes, it’s a femto clock. This eliminates any jitter from the incoming data stream as we reconstruct it with a new clock.
Hope this gives you some insight into the M21. It’s quite a complex product and a lot was done to get all 3 DACs working in a seamless manner.
Q: Are all the conversion process you mentioned performed inside a custom chipset, or is there a peripheral circuit? Also, for the very ubiquitous and cheap DSD chips you mentioned, what is one of the most deceptive processes possibly actually taking place?
Brian: All typical audio chips are in the low-price range whereas the expensive ones are $15-20, and that is a high price for a stereo audio dac chip. They are not DSD chips but multi bit PCM converters that accept DSD at the input and simply process for conversion. I don’t know of a real DSD chip, otherwise we would have not gone to the effort to design one. The Native DSD conversion inside the M21 is done via discreet analog circuits and all done in the analog domain.
To understand what happens in the audio DAC chips, you can’t make a real 1 bit converter unless you incorporate all we did in the chip in analog, but what happens is, it’s far easier to do it in digital and convert the data stream, filter it to a multi bit PCM stream, then you can perform DSP to solve these issues of DSD. There is one well known DAC that resamples DSD to SRCs and runs everything to a fixed PCM rate and takes in any rate you give it so yes, it converts the incoming data stream to analog audio.
Q: Tell me more about the construction of the DSD converter. What are the main technological highlights of it?
Brian: This starts a discussion and basic understanding of the problems with DSD. DSD one bit goes very fast in the case of DSD 64 at 2.8 MHz, so regardless of the rate, DSD 64 is 128 256, and 1 x 1 = 1 so there is no possibility to do any signal processing on it, none. Hence, all the chips and ladder dacs will have to convert it into PCM.
Our idea of the Native DSD circuit started in the M12 and carried over to the M21, for the purpose of making a real DSD converter; and if we are to make a true DSD converter we have to do it all in the analog domain. On the face of it, this looks quite simple. We use a 10GHz flip flop, which is a switch that will vibrate at the rate of the data stream, creating the modulation and the incoming Mclock from our DSS clock, which is followed by a gentle analog noise filter. After all, it’s one bit so it is on or off, or 2 states. That’s the easy part. Now you have a problem noise and with start and stop of the data stream. There is a filter after the switch, this is done in the analog domain and is a low order low pass filter, which is needed to remove noise and yield a low distortion signal. We then follow the filter with a Burr Brown level control to manage the fade in and fade out, mute, etc, and using this method with a sophisticated level controller allowing for a seamless integration with the PCM dacs. This way, we can create fast fades or hard mutes and control the level of the signal. To my knowledge this is one of the only real, true DSD converters in the market and we feel one of the great benefits of the M21.
In the end, when you add up all the components needed it is not a low cost method, it’s a discreet design and one must consider that the DAC chips everyone uses are in the $5-15 price range and they will tell you they accept any rate DSD, ah but what goes on inside is a mystery.
BTW, the real reason for the analog level control in the M21 was to manage the DSD, and it does become very handy to use the M21 direct to an amp.
Q: Regarding the NDSD PCM vs NDSD DSD settings, is there a situation in which the DSD setting would route non-DSD files to the custom DSD circuit nonetheless? Does it mean that custom circuit also process non-DSD files? In addition, I’d like to understand your choice of the Analog Device 1955 chips.
Brian: Only DSD sources use the NDSD converter when it’s set to NDSD. If the setting is changed from DSD to PCM, the DAC will use the AD 1955 for conversion.
All PCM sources use the AD1955 or Ladder DAC, so setting to Sigma will use the 1955 regardless of the NDSD setting. There is no upsampling of PCM to DSD to then use the NDSD converter, we feel there is no point to doing such processes as the AD1955 does an outstanding job of converting all PCM sources. The NDSD converter was made as a way to directly convert DSD masters and releases natively with no additional processing.
The AD1955 is a very fine part, its features allow us to do all processing outside of the chip. It allows for us to create the oversampling filter in our DSP, and that allows the part to run at 384k PCM (not 192 as in the chip manufacturer’s data sheet). We supply the I-V or current-to-voltage conversion and we do it with a very high speed fast settling opamp. All we use it for is the final sigma delta modulation. It has excellent distortion characteristics and this actually is more important than noise, but the noise too is extremely low. We typically get 0.0007% THD+N in test and 0.0004% THD. This part does not have sample rate converters in it as do some others, and can handle all common rates of PCM sources, from 44.1-384k.
Q: What is so special about the M21’s analog volume control system?
Brian: The M21 has an analog level control as this is needed to manage the NDSD, so all the mutes and fades are done in analog in the M21. It is a Burr Brown volume control chip but we implement it as a current-to-voltage device, and as such has outstanding performance. In most DAC including M1, this can be done in the digital domain as everything is converted to some form of PCM or multi bit. But if one were to have a pure 1 bit converter then all processing must be done in analog, including any filters. Therefore, the M21 has an analog volume control that is used for this purpose so we can deliver a true DSD playback with seamless presentation and with PCM playback as well. With the analog attenuator, the M21 has the added benefit of driving the amps directly with a preamp function as opposed to the digital attenuator in the M1. When the level control of the M21 is set to 0dB, there is a hard bypass of the attenuator when playing or streaming data, and only in the transitions does the attenuator come into play. This way, the M21 can be used as a pure DAC with a preamp of choice or used direct to the amps.
When using the M21 direct to the amps you have an analog attenuator and up to 4.5V on the balanced outs and about 3V on the unbalanced, and considering it’s an analog attenuator in the M21 there is no need to do anything, it’s set in a good range and you just use the analog attenuator like a preamp to control the level. We run the M21 direct to the M28 at shows and recommend that you try it that way.
Bricasti Design CEO Brian Zolner opens up on the company’s flagship product, the M21 dual-mono DSD DAC.
As I am about to publish my findings on the Bricasti Design M21 DSD DAC, the complexity of the $21k DAC is such that Brian Zolner, CEO of the company, and I conducted many correspondences on the subject. The extent and detail of our discussion is such that I thought it would be a good read for all to enjoy before the review’s publishing. The following is a condensed version of our discussion.
Q: How does the M21 improve upon the M1?
Brian: The M21 has the same discreet output buffer found in the M1, but it goes a bit further. One way to think about the M21 is an extension or upgrade of the basic M1 design, by adding the Native DSD converter that we first designed for the M12 Source Controller, in a product that can be used as a DAC and not a pre/dac. Or one could also say it’s like the M12 without analog inputs. This would not fit in the M1 chassis, so we took the larger chassis from the M12, used superior analog power supplies, making it more powerful than the M1, and then added the ladder dac just because we wanted to see how it would perform and compare with our M1 sigma delta dac. So, there are 3 DACs in the M21.
Q: Please explain the significance of the three DACs.
Brian: For DSD conversion, we made a true DSD converter in the M21. It was first made for the M12 Source Controller, and part of the reason for making the M21 was to offer this Native DSD converter in a product that could be used as a DAC where as the M12 is a pre/dac product. This is a true DSD converter and is based on a very high speed switch followed by an analog filter; any mutes and level adjustments for transition noise removal is done in the analog domain. When a DSD signal is converted in the M21, there is no downsampling to PCM as in all other types of converters. DSD presents issues if not downsampled for use in most chips as it is not possible to perform any signal processing on it, so even a start and stop of the data stream presents issues for other DACs, like snaps, pops at full scale output. Most DACs convert the signal to PCM and they are then able to remove any issues like this in the digital domain. Bricasti Design does this all in the analog domain and this allows for seamless transitions between DSD and PCM streams. This is why the M21 has an analog attenuator, the main purpose of which is to elegantly manage the DSD issues noted above and integrate it with the data streams from the PCM DACs.
For PCM processing, there are 2 PCM converters in the M21.
One is a sigma delta type based on the ADI 1955, which is the same as used in the M1 and has been our standard DAC design in all our products starting with the M1. We only use this part for its final upsampling to 5-bit DSD or otherwise known as the sigma delta modulator, and we create the filters in our DSP and our own I-V or current-to-voltage conversion with high speed opamps. This part of the design is taken from the M1, and in here, there is no need to deal with transition noise issues as the DAC can manage things in addition to the DSP.
The third one is a Ladder DAC. This is based on a 20-bit ADI part, and again we create our anti-aliasing filters in our DSP and analog filtering, enabling the DAC to manage any transition issues in DSP or with the analog stage of the M21.
An important part of the product is the way we do our clocking and clock recovery: whatever part or design for the modulation one uses, a key factor is the clocking. We use our DDS (direct digital synthesis) technique to reclock the incoming data stream. We track the incoming Mclock and buffer and resync against our DDS clock. Yes, it’s a femto clock. This eliminates any jitter from the incoming data stream as we reconstruct it with a new clock.
Hope this gives you some insight into the M21. It’s quite a complex product and a lot was done to get all 3 DACs working in a seamless manner.
Q: Are all the conversion process you mentioned performed inside a custom chipset, or is there a peripheral circuit? Also, for the very ubiquitous and cheap DSD chips you mentioned, what is one of the most deceptive processes possibly actually taking place?
Brian: All typical audio chips are in the low-price range whereas the expensive ones are $15-20, and that is a high price for a stereo audio dac chip. They are not DSD chips but multi bit PCM converters that accept DSD at the input and simply process for conversion. I don’t know of a real DSD chip, otherwise we would have not gone to the effort to design one. The Native DSD conversion inside the M21 is done via discreet analog circuits and all done in the analog domain.
To understand what happens in the audio DAC chips, you can’t make a real 1 bit converter unless you incorporate all we did in the chip in analog, but what happens is, it’s far easier to do it in digital and convert the data stream, filter it to a multi bit PCM stream, then you can perform DSP to solve these issues of DSD. There is one well known DAC that resamples DSD to SRCs and runs everything to a fixed PCM rate and takes in any rate you give it so yes, it converts the incoming data stream to analog audio.
Q: Tell me more about the construction of the DSD converter. What are the main technological highlights of it?
Brian: This starts a discussion and basic understanding of the problems with DSD. DSD one bit goes very fast in the case of DSD 64 at 2.8 MHz, so regardless of the rate, DSD 64 is 128 256, and 1 x 1 = 1 so there is no possibility to do any signal processing on it, none. Hence, all the chips and ladder dacs will have to convert it into PCM.
Our idea of the Native DSD circuit started in the M12 and carried over to the M21, for the purpose of making a real DSD converter; and if we are to make a true DSD converter we have to do it all in the analog domain. On the face of it, this looks quite simple. We use a 10GHz flip flop, which is a switch that will vibrate at the rate of the data stream, creating the modulation and the incoming Mclock from our DSS clock, which is followed by a gentle analog noise filter. After all, it’s one bit so it is on or off, or 2 states. That’s the easy part. Now you have a problem noise and with start and stop of the data stream. There is a filter after the switch, this is done in the analog domain and is a low order low pass filter, which is needed to remove noise and yield a low distortion signal. We then follow the filter with a Burr Brown level control to manage the fade in and fade out, mute, etc, and using this method with a sophisticated level controller allowing for a seamless integration with the PCM dacs. This way, we can create fast fades or hard mutes and control the level of the signal. To my knowledge this is one of the only real, true DSD converters in the market and we feel one of the great benefits of the M21.
In the end, when you add up all the components needed it is not a low cost method, it’s a discreet design and one must consider that the DAC chips everyone uses are in the $5-15 price range and they will tell you they accept any rate DSD, ah but what goes on inside is a mystery.
BTW, the real reason for the analog level control in the M21 was to manage the DSD, and it does become very handy to use the M21 direct to an amp.
Q: Regarding the NDSD PCM vs NDSD DSD settings, is there a situation in which the DSD setting would route non-DSD files to the custom DSD circuit nonetheless? Does it mean that custom circuit also process non-DSD files? In addition, I’d like to understand your choice of the Analog Device 1955 chips.
Brian: Only DSD sources use the NDSD converter when it’s set to NDSD. If the setting is changed from DSD to PCM, the DAC will use the AD 1955 for conversion.
All PCM sources use the AD1955 or Ladder DAC, so setting to Sigma will use the 1955 regardless of the NDSD setting. There is no upsampling of PCM to DSD to then use the NDSD converter, we feel there is no point to doing such processes as the AD1955 does an outstanding job of converting all PCM sources. The NDSD converter was made as a way to directly convert DSD masters and releases natively with no additional processing.
The AD1955 is a very fine part, its features allow us to do all processing outside of the chip. It allows for us to create the oversampling filter in our DSP, and that allows the part to run at 384k PCM (not 192 as in the chip manufacturer’s data sheet). We supply the I-V or current-to-voltage conversion and we do it with a very high speed fast settling opamp. All we use it for is the final sigma delta modulation. It has excellent distortion characteristics and this actually is more important than noise, but the noise too is extremely low. We typically get 0.0007% THD+N in test and 0.0004% THD. This part does not have sample rate converters in it as do some others, and can handle all common rates of PCM sources, from 44.1-384k.
Q: What is so special about the M21’s analog volume control system?
Brian: The M21 has an analog level control as this is needed to manage the NDSD, so all the mutes and fades are done in analog in the M21. It is a Burr Brown volume control chip but we implement it as a current-to-voltage device, and as such has outstanding performance. In most DAC including M1, this can be done in the digital domain as everything is converted to some form of PCM or multi bit. But if one were to have a pure 1 bit converter then all processing must be done in analog, including any filters. Therefore, the M21 has an analog volume control that is used for this purpose so we can deliver a true DSD playback with seamless presentation and with PCM playback as well. With the analog attenuator, the M21 has the added benefit of driving the amps directly with a preamp function as opposed to the digital attenuator in the M1. When the level control of the M21 is set to 0dB, there is a hard bypass of the attenuator when playing or streaming data, and only in the transitions does the attenuator come into play. This way, the M21 can be used as a pure DAC with a preamp of choice or used direct to the amps.
When using the M21 direct to the amps you have an analog attenuator and up to 4.5V on the balanced outs and about 3V on the unbalanced, and considering it’s an analog attenuator in the M21 there is no need to do anything, it’s set in a good range and you just use the analog attenuator like a preamp to control the level. We run the M21 direct to the M28 at shows and recommend that you try it that way.
NEXT: Bricasti Design M21 dual-mono DSD, Ladder and Delta Sigma DAC Review.