"DSD Duel: Class A Video Amplifiers vs. CIC Filters – Who Wins the Hi-Fi Battle? PSAUDIO VS SIGNALYST"

There are many ways to 'skin a cat' as we say down here in the heart of Appalachia, with the Smoky Mountains and 'Rocky Top' in direct view from my porch as I am writing this.  A truly inspiring scene to relax and then paradoxically write a mind numbing technical comparison. 

Why is DSD on my mind?  I seem to have a sickness; a truly impulsive need to explore its 'mysteries'.

Signalyst Discrete DSD DAC PCB

PSAudio DirectStream DAC Internals

Today I have chosen two 'Direct' DSD DACs to compare their approach to Digital to Analog Conversion. 

The first of these is the Signalyst DSC Discrete DSD DAC, which cannot be bought via any normal 'retail' method.  You have to build it yourself or have one built for you.  The intellectual creator is Jussi Laako of Signalyst, maker of ​HQPlayer Software. My actual build is the work of Pavel Pogodin.  You can read more by clicking here.  Before going any further, I need to point out this is a  DSD only DAC.  PCM cannot be played without conversion to DSD.  The DSC DAC is designed to be used with HQPlayer software, a powerful tool that converts PCM to DSD, and lower rate DSD to higher rates up to DSD1024.  (Note the actual DSC DAC can only accept up to DSD512.)  

The second of these DSD DACs under discussion is the PSAudio DirectStream DAC, which comes in two versions, MK1 or MK2.  The distinctions between MK1 and MK2 play no real role here; the basics are the same.  The DIRECTSTREAM will accept both PCM and DSD, and needs no external software for PCM because similar conversion functions are done internally in the FPGA brain of the DAC.  

Due to the 'separates' nature of the HQPlayer external software combined with the DSC DAC, you actually get what may, counterintuitively at first glance, be an advantage.  Although both DACs boast 'Direct' DSD processing, only the DSC DAC can actually convert DSD with no extra DSP required, because the PSAUDIO DAC has no type of DSD 'bypass' mode.  DSD via the PSAUDIO DIRECTSTEAM will always undergo DSP and is never 1-bit at all times.   

The SIGNALYST DSC DAC can actually process DSD directly with no DSP via two methods.  The first is simply connect the DAC to a PC, Network Streamer with USB output, etc., and only send it DSD signals.  The other way is to use flexible software, such as the HQPlayer software, in which the user has a choice to send DSD files directly and bitperfectly to the DAC as an additional possible path if one wishes not to use any DSP.   However, it is probably necessary to point out that the DSP systems in both HQPLAYER and in the DIRECTSTREAM (especially so in the DIRECTSTREAM) are programmed for optimal synergy with their respective hardware analog conversion systems.  

I do not want to delve too much more into the DSP side of things, because I want to focus here on the actual conversion of DSD itself, after any DSP.  But the reason the DSP exists is to provide extremely well designed oversampling filters for both PCM and DSD, which can be more capable and accurate than what you find in a typical DAC.  It also allows for digital volume control, even on DSD.  

Once things get past the DSP, things really get a bit more similar.  Both use discrete analog components to filter their DSD signal (Reminder, no matter what signal you put in either system, either PCM or DSD, it will be internally converted to DSD for analog conversion.)  The DIRECTSTREAM claims to be a completely passive system; however the SIGNALYST DSC uses a Sallen-Key Filter as part of its DSD filtering, which is an active component.  

So, how do they work??  Here comes the fun part.  Or the part where many of you may tune out.  Technical stuff is ahead.  You have had your fair warning :) 

---

Let's start with the Signalyst. 

It uses what I would call a more traditional and common method of DSD conversion.  Most DACs, even though they are not made of as many discrete components, use something similar IF they have a bypass or native DSD mode.  

Disclaimer.  These descriptions are a bit generalized.  They may not take into account every possible step or piece of hardware.  

The SIGNALYST DSC starts by receiving the 1 bit DSD signal, which can be as high as DSD512, and sends it into a discretely built analog filter.  The filter uses typical digital techniques, but is implemented in the analog domain.  Some may see this as a Digital/Analog Hybrid filter.  The filter is a type of CIC filter, with FIR filter characteristics, that excludes the decimation stage.  It just filters and smooths samples into analog.  It doesn't discard any actual samples in the process.  

The way this works is pretty darn ingenious.  What you need are shift registers (flip-flops, no not the sandal), a MOSFET or transistor to act as a switch that either connects or disconnects an 'output element' resistor (which is the equivalent of a digital filter TAP), and some kind of summation node. 

CIC FILTER

1. Shift Register:
   • Function: The shift register takes the incoming DSD bitstream and creates multiple parallel bitstreams, each offset by one clock cycle. In the case of the SIGNALYST DSC DAC, it takes 32 consecutive bits of the DSD stream, and places the 32 identical DSD bitstreams stacked upon one another, and remember each stream is offset by one tick of the bit-clock.  Due to the nature of this kind of bit coding (thermometer code), it actually means 33 conversion levels, because we can't forget about 0 level.  The timing offset out of the shift register creates a smoothing, comb filter effect when combined with the output elements (voltage controlled resistors) and final summation of all signals into one again.      
2. Voltage-Controlled Resistors and Switches:
   • Role: Each bitstream controls a switch (e.g., a MOSFET or a transistor) that either connects or disconnects a resistor from the summation node.
   • Design: The resistors can be chosen to provide a weighted contribution to the summation based on the timing of the bitstream. In the case of the DSC DAC, however, the element are of equal weighting, so there is no further contribution to the filtering.  
3. Analog Summation:
   • Summation Network: The analog outputs of the switches are summed together. This summation integrates the 1-bit DSD bitstream into a continuous analog signal.
   • Filtering: The combined effect of the time-offset bitstreams, output elements, and the summing network completes the low-pass filter, smoothing the high-frequency components and leaving a clean analog signal.

But this still isn't quite enough filtering for the extremely high levels of ultrasonic noise.  It DID accomplish conversion from digital to analog, and did a great deal of filtering itself, but we need more.  

The SIGNALYST DSC DAC follows the CIC Comb filter that converted the digital signal to analog with another analog filter.  One that assists in further shaping out that ultrasonic noise.  In comes some active filtering.. a Sallen-Key filter. 

SALLEN-KEY FILTER

• Additional Filtration: The Sallen-Key filter will provide further low-pass filtering to ensure that any high-frequency artifacts not completely attenuated by the passive filter are removed.
• Signal Conditioning: It can also help in signal conditioning, providing a clean and smooth analog output.  (It is worth noting that some other similar designs stay passive with an RC filter in this position instead. (iFI Audio, I am talking about you-- also note iFi while using a very similar conversion technique, uses unequally weighted elements and has a bitstream that is only 8 bits long.)  

By following the passive discrete component hybrid digital/analog CIC output filter with an active Sallen-Key filter, we achieve a robust and comprehensive filtering solution for converting DSD to a high-quality analog signal. This combination leverages the strengths of both passive and active filtering techniques, ensuring minimal high-frequency noise and excellent signal integrity.

FINAL STAGE: OUTPUT TRANSFORMER

While most other DACS I can think of use different kinds of final analog output methods, both the SIGNALYST DSC DAC and the DIRECTSTREAM DAC have chosen to use output transformers.  

Purpose:

• Impedance Matching: Ensures the output impedance matches the input impedance of the next stage (e.g., an amplifier or audio interface).
• Isolation: Provides galvanic isolation to reduce ground loops and noise
• Signal Smoothing: Further smooths the signal by filtering out any remaining high-frequency components.

By following this design approach, the SIGNALYST DSC DSD DAC achieves high-fidelity conversion of DSD to analog, leveraging the strengths of discrete components and innovative filtering techniques.  One thing to note before we move on to the DIRECTSTREAM, is the CIC filter by its natural design will change its filter cutoff frequency with each change of DSD speed, and it will double with the change.  DSD64 may hypothetically start its rolloff at 30khz.  DSD128 would start at 60khz, DSD256 would start at 120khz, etc.  This is important for later comparison with the DIRECTSTREAM DAC.

Moving on to the PSAudio DIRECTSTREAM DAC

This one is unique.  I know of no other DAC currently available that uses this technique.  The previously discussed technique used in the SIGNALYST DSC DAC is quite standard across the industry, and the schematics for it are Open Source, so its easy to get to the details of operation. Not so here.  We have some major differences, derived from a few clues thrown our way.  Because it's proprietary intellectual property, expect a shorter and less deep dive into its operation. 

The DIRECTSTREAM, as mentioned in the beginning, contains its own bespoke digital filters and digital volume control on its FPGA.  The 'intermediate signal' where the Volume Control, Balance Control, and whatever other DSP it uses, is at least at 30bit per sample signal at least 10x the DSD64 rate.  This minimum 30bit ultra-high sample rate signal is used for DSP on both PCM and DSD.  

They advertise this is always a 1-bit system that never is converted to PCM. ​ I find this misleading and inaccurate.  What they are trying to say is, when a 1-bit PURE DSD SIGNAL is input into the DAC, the signal isn't ever decimated to any type of low PCM rate.  The truth is, both PCM and DSD use an interpolation filter.  Once DSD is interpolated, it is no longer a 1-bit, time splicing noise shaped signal, although of course this 'intermediate' signal can be oversampled and re-noise shaped into whatever bit depth and sample rate one could want.  

The actual DSD signal is oversampled by probably an FIR filter, just like Sony DSD-Wide of old, and ESS Sabre of today, into a huge 30 bit 28.224 MHZ signal!! (MKI) 

(Nothing new is under the sun, and there is no 'magic' in how DSP is applied to 1-bit DSD.  Even now, decades later, the best DSD recording systems are using the same techniques as yesteryear.  Many  are seeming to stay with a Sony 'DSD-Wide' type approach.)

Why a signal so big?  Well, one reason would be you can use tremendously large digital FIR filters with extreme accuracy and control, by being able to implement millions upon millions of filter TAPS.  This is evidenced by the impulse response measurements that have appeared in the big pro magazines when the PSAUDIO DIRECTSTREAM DAC is on their test bench.  The filter rings seemingly forever, reminding me of a Chord product.  Yes, there are advantages here, but, all that ringing is a major disadvantage.  But that is a subject for a different day.  Additionally, with DSD material, the FIR oversampling filter could help with ultrasonic noise control as it will pre-filter the signal in this multi-bit stage.  

After the DSP is finished, it uses a delta-sigma modulator to convert everything to DSD128.  Remember how the SIGNALYST DSC outputs multiple rates that change the filter characteristics?  That doesn't happen here.  Everything is converted, PCM and  DSD no matter what the rate, even DSD 256, to DSD128.  WHY?  That is something more than this article can cover, but there is the idea of a DSD 'sweet spot' where extra speed is actually detrimental to the sound and makes for a more difficult analog conversion, counterintuitively at first, until you understand why.  I suggest you read Andreas Koch talk about it here.  

So we make it to our final bitstream. 1 bit DSD128.  This is where the fun begins (again)....

Instead of the more common discrete CIC (FIR) filters used to convert 1 bit DSD to analog, the DIRECTSTREAM uses something I would never have thought of... a class A video amplifier!!!

Using a Class A video amplifier to filter a DSD bitstream is a quite sophisticated method to achieve high-fidelity audio output. This approach leverages the high-speed and wide bandwidth capabilities of video amplifiers, which can handle the high-frequency components of DSD signals effectively.

CHARACTERISTICS

1. High Bandwidth:
   • Video amplifiers typically have very high bandwidth, well into the MHz range, which is essential for handling the high-frequency content in DSD signals (e.g., DSD64 at 2.8224 MHz).
2. Low Distortion:
   • Class A operation ensures low distortion and high linearity, which is critical for maintaining the integrity of the audio signal.
3. High Slew Rate:
   • The ability to respond quickly to changes in the signal makes video amplifiers suitable for the fast transitions present in DSD bitstreams.

MORE CONCEPTUALIZATION

1. Direct Amplification:
   • The DSD bitstream is directly fed into the Class A video amplifier. The amplifier’s particular bandwidth allows it to pass the high-frequency components that are to be kept, and to attenuate those that need to be discarded.  
2. Low-Pass Filtering:
   • By leveraging these frequency response characteristics of the amplifier and possibly additional passive components, high-frequency noise can be filtered out, leaving a clean analog signal.

CIRCUIT DESIGN

1. Input Stage:
   • The input stage receives the DSD bitstream and prepares it for amplification. This stage needs to be designed to match the impedance of the bitstream source and ensure proper signal levels for the amplifier.
2. Amplifier Stage:
   • A high-bandwidth Class A video amplifier must be used. These amplifiers provide the necessary speed and linearity.  The output of the amplifier stage is fully analog.  The digital bitstream has now been converted into a filtered higher voltage analog representation.  
3. Output Filtering:
   • An RC (resistor-capacitor) network can be used at the output to provide additional low-pass filtering. This network can help to further smooth the signal by attenuating frequencies above the audible range.

ADVANTAGES

1. High Fidelity:
   • The use of Class A video amplifiers ensures high fidelity due to low distortion and high linearity.
2. Wide Bandwidth:
   • Capable of handling the high frequencies associated with DSD bitstreams.
3. Simplicity:
   • Simplifies the filtering process by leveraging the inherent characteristics of the video amplifier.

We are not finished yet.  Just like the SIGNALYST DSC DAC, the DIRECTSTREAM uses an output transformer for the same exact functions.  It offer some filtration to go along with the filtration of the Video Amplifier, along with possible other passive analog filtering such as an RC filter.  Often this  output transformer filter function is referred to as working at DSD256. 

That made no sense to me at first.  I first saw it stated that way in Hi-Fi News.  But, the transformer doesn't put out "DSD256" by any means, nor any other bitstream.  It is an analog signal at this point.   

What is happening here is this: the previously discussed full analog system (sans transformer) is designed for conversion and filtering of one rate: DSD128.  But the TRANSFORMER is optimized for DSD256 filtering.  Here is why:

​If the output transformer, which has its own low pass filter capabilities, is optimized for DSD256 and is used with a DSD128 signal, it will have a higher cutoff frequency. This approach helps to preserve the transient response of the signal. Here’s a detailed explanation of why this is the case and the implications for audio performance:

Output Filter Optimization

1. Filter Cutoff Frequency:
   • DSD128 vs. DSD256: DSD128 has a sampling rate of 5.6448 MHz, while DSD256 has a sampling rate of 11.2896 MHz. A filter optimized for DSD256 would typically have a cutoff frequency that is suitable for handling the higher frequency noise components associated with the higher sampling rate.
   • Higher Cutoff Frequency: When this filter is applied to a DSD128 signal, the higher cutoff frequency allows more high-frequency content to pass through, which can improve the transient response of the audio signal.
2. Transient Response:
   • Preservation of High-Frequency Details: A higher cutoff frequency means that more high-frequency transients and details are preserved in the analog output. This is crucial for maintaining the clarity and accuracy of fast, transient-rich audio signals.

Practical Implications

1. Noise Shaping and Filtering:
   • High-Frequency Noise: DSD signals inherently contain high-frequency quantization noise. Filters optimized for higher rates (like DSD256) are designed to attenuate this noise effectively without impacting the audible range.
   • Application to DSD128: When such a filter is applied to a DSD128 signal, it will still attenuate high-frequency noise but may do so less aggressively than a filter specifically designed for DSD128. This results in a cleaner transient response but may allow some high-frequency noise to remain.
2. Audio Quality:
   • Enhanced Detail and Clarity: By preserving more high-frequency content, the audio output can benefit from enhanced detail and clarity, particularly in complex and dynamic recordings.
   • Potential Trade-offs: There is a balance to be struck between preserving transient response and minimizing high-frequency noise. The design of the filter must consider this to optimize overall audio performance.
   •  The filter designed for DSD256 will attenuate less aggressively, preserving more high-frequency content and improving transient response, at the cost of potentially letting through more high-frequency noise.

Optimizing an output filter for DSD256 and using it for a DSD128 signal results in a higher cutoff frequency, which helps in preserving transient response. This approach enhances the clarity and detail of the audio signal but must be carefully balanced to manage high-frequency noise effectively. This method highlights the importance of considering the specific characteristics of both the signal and the filter in high-fidelity audio design.



So now let's look at a direct comparison between the SIGNALYST DSC DAC and the PSAUDIO DIRECTSTREAM DAC, highlighting relative strengths and weaknesses. 

Method 1: Class A Video Amplifiers for Filtering DSD Bitstream
Characteristics:

• High Bandwidth: Class A video amplifiers can handle the high-frequency content of DSD signals effectively.
• Low Distortion: Class A operation ensures low distortion and high linearity.
• High Slew Rate: Suitable for fast transitions in DSD bitstreams.

Implementation:

• Direct Amplification: The DSD bitstream is directly fed into the Class A video amplifier.
• Output Filtering: Possibly uses RC networks for additional low-pass filtering to remove high-frequency noise.
• Output Transformer: Ensures impedance matching, signal isolation, and additional smoothing.

Advantages:

• Simplicity: Direct approach with fewer components.
• High Fidelity: Low distortion and high linearity provide excellent audio quality.
• Wide Bandwidth: Capable of handling high-frequency components inherent in DSD.

Disadvantages:

• Power Consumption: Class A amplifiers dissipate a lot of power and require good thermal management.
• Limited Filtering: While video amplifiers have wide bandwidth, they may not provide as precise filtering as dedicated filter circuits.

Method 2: Discrete Component CIC Filter Plus Sallen-Key Filter and Output Transformer

Characteristics:

• CIC Filter: Uses shift registers and resistor networks to create a composite analog signal from the DSD bitstream.
• Sallen-Key Filter: Provides precise low-pass filtering to remove high-frequency noise.
• Output Transformer: Ensures impedance matching, signal isolation, and additional smoothing.

Implementation:

• CIC Filter: Shift registers create multiple parallel bitstreams, each processed through resistors and then summed.
• Sallen-Key Filter: Active low-pass filter with high precision.
• Output Transformer: Final stage for signal conditioning and isolation.

Advantages:

• Precision: Sallen-Key filters provide precise control over the filtering characteristics.
• Comprehensive Filtering: Combined stages ensure thorough removal of high-frequency noise.
• Signal Conditioning: Output transformer adds benefits of impedance matching and isolation.

Disadvantages:

• Complexity: More components and stages involved.
• Size and Cost: Potentially larger and more expensive due to the number of components. (from a raw parts perspective- not a retail perspective)
• Active Components: Requires power and careful design of active filter stages.

COMPARATIVE ANALYSIS

1. Filtering Precision:
   • Class A Video Amplifiers: Good for general filtering with high fidelity but may not achieve the same level of precision in filtering high-frequency noise as the discrete component approach.
   • Discrete Component Approach: Offers more precise and controlled filtering, particularly effective in removing high-frequency noise due to the combination of CIC and Sallen-Key filters.
2. Complexity and Power Consumption:
   • Class A Video Amplifiers: Simpler with fewer components but higher power consumption and heat dissipation requirements.
   • Discrete Component Approach: More complex and larger, with additional power requirements for active components, but generally more efficient in specific filtering tasks.
3. Audio Fidelity:
   • Both methods can achieve high audio fidelity, but the discrete component approach with Sallen-Key filtering might offer better overall noise reduction, especially for high-end audio applications where precise filtering is critical.
4. Implementation and Cost:
   • Class A Video Amplifiers: Easier to implement with fewer components, potentially lower cost for simpler designs.  (not from a retail perspective)
   • Discrete Component Approach: Higher complexity and cost but provides a more comprehensive filtering solution. (not from a retail perspective)

CONCLUSION
The choice between using Class A video amplifiers or a discrete component CIC filter plus Sallen-Key filter with an output transformer depends on the specific requirements of the application:

• For simplicity and high fidelity with high bandwidth capability: Class A video amplifiers are a good choice, especially if power consumption and thermal management can be handled.
• For precise and comprehensive filtering: The discrete component approach with CIC and Sallen-Key filters, followed by an output transformer, offers superior noise reduction and signal conditioning, making it ideal for high-end audio applications.