I wanted to build a server that is capable of running multichannel pipeline matrix processing (including convolution) and upsampling to the 8-channel exaSound e28 DAC at DSD256.
One challenge was finding a nice CPU for the task. When the new Broadwell-E generation of CPUs became available the choice became easier. Either 8-core i7 6900K with 3.2 GHz base frequency, or 10-core i7 6950X with 3.0 GHz base frequency. I ended up selecting the 6950X because then there's a core per channel
I built a new low acoustic noise machine for using HQPlayer on Windows 10 with exaSound e28 DAC (8 channel). This machine is currently based on Gigabyte GA-Z170X-UD5 TH main board for their Ultra Durable series. It has USB 3.1-C / Thunderbolt among all kinds other very recent features. CPU is Core i5-6600T (35W TDP) and RAM is 32 GB (4 DIMMS) of Kingston HyperX FURY Black DDR4 2133 MHz CL14 and passively cooled Seasonic Platinum 520 Fanless PSU.
Since the drivers for Intel GPU built into the
Now since there are MQA encoded FLAC test files available, I wanted to see how the file works out without MQA decoder - as a normal FLAC.
If we first look at the original 352.8/24 (DXD) file that has FLAC size of 69 MB:
We can see that there is content up to around 57 kHz.
If we then convert it to 44.1/24 we get a FLAC with size of 14 MB:
From the filter roll-off we can see that there's quite a bit less than 24-bit worth of infor
There has been a lot of discussion about different types of volume controls. For comparison purposes of one case I chose to use Mytek Stereo192-DSD DAC which has three different volume control modes; analog, digital and bypass. Signal routing is implemented using relays. DAC chip is ESS Sabre ES9016S in stereo mode (4 DACs in parallel per channel). ESS Sabre has 32-bit volume control and internal DSP pipeline so it has 8 bits more resolution than allowed by 24-bit external input. It also has hig
Based on request here, I made some more square wave scope measurements using a high resolution scope for comparison with earlier measurements of DSC1.
Here are 7 kHz square wave results for the DSC1.
DSD128:
DSD256:
DSD512:
Here are 7 kHz square wave results for the iFi iDSD Micro.
PCM 352.8k:
PCM 705.6k:
Here are my standard set of measurements, this time with iFi iDSD Micro.
First sweep 0 - 22.05 kHz. Source is 44.1 kHz 32-bit PCM file.
Filter "Bitperfect" (NOS):
Filter "Standard":
Filter "Minimum(phase)":
Upsampled to 768/24 PCM in HQPlayer:
Upsampled to DSD256 in HQPlayer:
Upsampled to DSD512 in HQPlayer:
As requested by Superdad, here are sweep results for a 0 - 768 kHz sweep (source is 1536/32 PCM).
DSD128:
DSD256:
DSD512:
Note that corner frequency of the analog filter formed by the conversion stage moves up as the sampling rate is increased, while the Sallen-Key filter has static corner frequency.
Rise and stability of 7 kHz square wave at DSD256:
I finally used couple of days to do small fixes for the first unit I put together and got first decent measurement results.
Results are still considered preliminary, because the DAC was measured without any shielding case, plain boards and transformers on a table, connected with open unshielded wire hassle...
I performed my usual set of measurements.
First was 0 - 22.05 kHz linear frequency sweep (44.1 kHz 32-bit PCM source file) with analyzer set to 5
In response to endless DSD arguments here at CA I promised to publish an open hardware design of 1-bit multi-level DAC without magical black boxes. No DAC chips. No camel-back DSD ultrasonic noise bumps either, even with 7th order modulators. Multi-level, but still 1-bit. How is that possible? Well, it's not a secret and I wanted to put it out in an open way. A DAC that is not trying to play dual-role between it's PCM-to-SDM and DSD.
Moving most of the DSP things to software, having only min
Schiit Loki is an interesting product at very affordable price tag. At this price point, there's not much competition on desktop DAC market and even less (none?) on DSD DAC market.
Loki is based on Asahi Kasei Microdevices (AKM) AK4396 DAC chip and asynchronous UAC2 USB interface is implemented using C-Media CM6631A chip. The AK4396 has been one of the most popular DAC chips in pro-audio converters. Loki is USB-powered. Analog reconstruction filter is optimized for DSD use. Since COTS USB-ch
This is interesting DAC with a reasonable price tag, looks nice and high quality components.
Since it also offers bypass of the digital interpolation filters with direct 32-bit 352.8/384 PCM input as well as DSD128 support with good set of analog filters (conversion mappings), I wanted to study which would be best way to play back RedBook content on it.
I have excluded time-domain behavior inspections from this post, because going to that domain while keeping the frequenc
Here's now the promised update with 16-bit TPDF dither. This is also good for comparing how much dither vs truncation makes difference in real world.
CS4328 (1-bit 64x oversampling delta-sigma)
Updated numbers THD=0.002%, IMD=0.003%.
1 kHz sine:
<p><a href="/monthly_2013_07/CS4328-1k.png.df27de96ed3830ef16e548317e4c4dad.png" class="ipsAttachLink ipsAttachLink_image"><img data-fileid="28194" src="/monthly_2013_07/CS4328-1k.png.df27de96ed3830
I took a pair of my old early 90's DACs from the storage and fed bunch of measurements through both. Source for both was MuFi V-Link192 at 44.1k output. As result of doing this at midnight I forgot to change the output to 16-bit, so you may see some truncation quantization effects too... Both of the DACs claim 18-bit resolution.
Participants:
1) Crystal Semiconductor CS4328 64x oversampling 1-bit DAC chip with 8x interpolation and fifth-order modulator. IOW, so it's prett
