Gang,
Sorry for the delays it is my busy time.
To JohnW... I talked to TI about the clock issue there are actually 2 WCLK in the part. One is tracking the incoming stream and therefore moving. It determines the speed and from that MCLK and divided down for BCK and WCLK. The document is poorly translated to English from Japanese so the confusion.
I have also been trying to hammer out a more consistent testing system and so here that is. PC using Dr. Jordan's FFT software with ASIO and 64Bit math module this is the PRO version. I am using an EMU 0404 but I am trying to find a better card that allows 88.2 & 176.4 sampling as I think JTESTing at these sample rates will work better than others because of spectral leakage which essentially has the effect of "smearing" energy at one frequency to other frequencies in the spectrum.
I also picked up a PicoScope 12bit 10MHZ unit which has a pretty good FFT for measuring spectral of the clock signals.
All USB streams are originating from my MAC G5 dual processor. The computer has a 2.0 host controller with embedded Audio spec 1.1 and therefore no issues with the KMIXER.
Ok I have been working in my limited spare time on a new USB dac based on the 2706 & TDA1543. This is just a prototype strung out on my bench so measurements are so so. I have 2 boards they are the same except for the oscillator. The first is an ECS and the second the Low Jitter Cystek. The boards do have all the 3.3V signals tied to one LT1121-3.3 and bypassed as in the 2706 document. I will try and make another boad with seperated power supplies as time permits. All bypass caps are SMD as well as the resistors. The TDA1543 is running at 6V and uses leaded bias and passive I/V components. Main supply is a 9V supply from my bench linear supply.
The fist graph is the spectral of the ECS version of the OSC at BCK on the TDA1543.
This is with the low jitter Crystek oscillator @BCK:
These results differ from John's pretty much. The PicoScope can only do 4096 windows and the max sample rate is 10MHZ so maybe this is the cause of the differences or maybe because these use an OSC instead of crystals. I think much more work needs to be looked at in regards to the CLOCK inputs on the 2706.
Here is the JTEST results from both boards running at different windows and sampling rates and math models.
All are done at 24 bits input.
BLUE= 96K sampling, 64 bit math, 64K window ECS
WHITE= 44.1K sampling, 32 bit math, 64K window ECS
Gray = 96K sampling, 64 bit math 256K window ECS
Red = 96K sampling, 64 bit math, 256K window Crystek
I have been making amplifiers allot lately when I get a chance I will shot over some data from the commercial dacs I am making.
I still think a Lynx L22 maybe better for the JTEST in this regard.
Thanks
Gordon
Login
Register
FAQ
Search
!
! Had again problems with tantal caps, they started to draw current.