Two reasons:
1- Upsampling is used in most DACs out there to have better high frequency response - avoiding a steep brick wall filter results in better sound - this has been done for decades (recall the 8x oversampling CD players?)
2- It is a great way to justify a DAC specific licensing - kudos to their PR team for that one!
Iām gonna have to do this: Math is not engineering or psychological math, it is math and thereās no way to bend it.
āFoldingā a 96KHz/24bit signal into a 48KHz/24bit can only render something like a 96KHz/16bit signal out. Is this a real limiting factor? Quite unlikely, I would think the 16bit depth is probably more than enough for anything we have recorded (ie noise level is probably higher than that).
So I can accept the claim that most acoustic information is recorded in this equivalent 48KHz/24bit signal, I get that.
I can sum this up. 24/96k signal to MQA 24/48k is down-sampled, simple mathematics. First unfold MQA 24/48k gives 17/96kHz(MQA core decoding), trade in bit resolution to give higher sampling frequency. Final renderer, up-sample(if original recorded sample is greater than MQA core decoding) and impulse response filter optimisation. Thatās simple!
1 Like
ESS supports custom filters. Hereās what the DragonFly designer said:
There were no additions to the ESS DAC that are MQA specific. The firmware update did not update anything in the Dragonfly ESS DAC chip. Yes, it does support different upsampling filters. What the MQA decoding in the Dragonfly does it select those filters.
ESS has no custom filters, all available filters can be found in its data sheet. These are selectable hardware wired filters, not programmable unless you are talking about FPGA chip. The renderer is done prior to ESS DAC.
1 Like
But I found most of the MQA certified renderer/decoding DAC/DAP were ESS DAC based, I think MQA and ESS have deeper cooperation than we might think of, or ESS DAC born to be MQA ready by design.
I doubt that is the case for the current crop of ESS DAC chips.
ESS DACs are used in many higher end DACs, and the rendering code (which sets filters appropriately based on MQAās selection in the PCM data) is readily available, so I presume it is that combination that makes it easy for ESS DAC chip users to implement.
There might also be other architectural advantages such as common implementations of the chip using a field programmable controller rather than one that cannot be flashed, but I donāt know if this is the case.
That said, IF MQA really takes off, it is not unthinkable that DAC chip manufacturers will work with MQA to get better suited filters. I just donāt think thatās the case today.
Research into DAC filters and perhaps some collaboration would be a sensible first step. And it is likely MQA did that due diligence. Chips to volume produced hardware to the consumer. It would have been foolish to propose a new concept without figuring out if it could be done.
My understanding of the MQA filters is thereās nothing particularly special about them. Most DAC chips support min phase filters, which I understand is used here. There are many articles on this. Hereās one:
Among the dozens of MQA upsampling filters documented, many of the impulse responses are rather exotic ā some might even call them spurious. Though I would not jump to any editorial conclusions from that, the unusual impulse responses do suggest a degree of cooperation with MQA.
AJ
Ok thatās pretty interestingā¦ Butā¦ If you look at the ESS 9016 datasheet (the ESS DAC in the Dragonfly Red) it dates from Jan 17 2011. So those filters were largely determined at the time. There was no MQA then at all.
The Lumin Roon Ready MQA network player product line includes several models that use WM8741, so this proves that MQA can work on non-ESS DAC. Likewise, PS Audio MQA implementation resides in the Bridge II module only and requires no change from their DAC (based on my understanding of their public posts), so itās another proof.
However, Iād not disagree that ESS is suitable for MQA because ESS supports custom filters. (There are also other reasons as well.)
All MQA Core and Renderer are normally done in the XMOS chip and output via I2C bus to a DAC chip directly. Most of commercial off selves DAC can have its over-sampling digital filter completely switch off(via a command signal from XMOS chipset) and use the renderer instead.
Of course, it can still be process by a low powered CPU but the CPU output via I2C bus must be connected to a DAC chip directly. This way the CPU will tell the DAC to disengage itās over-sampling digital filter and use the renderer instead. Thereās NO custom MQA impulse filters or programable in those off selves DAC unless stated in their data sheets! Unless you are taking on FPGA discrete DAC.
Not in the Dragonfly. The controller receives the first unfolded PCM stream, reads the MQA filter selection, and sets that filter in the ESS DAC. That is all rendering is doing in that case.
1 Like
And making the light purple! Crucial! 
1 Like
Given that the filter can change, this is a good thing ā but it requires that you trust the MQA encoderās and/or teamās choices. There is nothing magic about the rendering ā but it is supposedly better than generic upsampling.
All jest aside, the Authentication in MQA is a great feature in a world of streaming services and pirates, where changes like this often happen (via malice and via incompetence). Think about all the people listening to AAC on TIDAL before signal path! Think about the number of FLAC files in the pirate scene that are actually reencoded 320k MP3.
This applies less so to labels/distributors, since they are the gatekeeps of authentication, so itās not a win-in-every-case situation, but itās better than the crap happening in the streaming/piracy scenes.
2 Likes
Meridian using apodizing minimum phase filter as default, Wolfson also included Min phase filter as one of the filter options since WM8741, so MQA should be working well with DAC chip with Minimum Phase Filter
Min phase filters are not a new thing! 
Well you always have to trust those implementations. I think having upsampling filter choices determined by the PCM stream is actually very smart and a good addition. But in most cases there is little advantage of doing so in the DAC vs using software upsampling with the specified filter transfer function to the max rate the DAC can take.
One exception to the above is actually the Dragonfly: max input rate is 96KHz, but max rate the ESS DAC can deliver is 384 KHz, so effectively the Dragonfly behaves just as effectively as a 384KHz DAC when playing MQA. Talk about a perfect fitā¦
As for the light, I completely agree with you. Especially in the case of many tracks I have bought in āhigh res 24/192ā only to run them through Musicscope and discover it is good-ole 44KHz upsampled. I truly hope this is never the case with MQA authenticated streams.
1 Like