Re: LF: Re: VLF Stability and soundcard locking

[Andy Talbot <andy.g4jnt@gmail.com>]

That is very much what I feared would happen.

 

USB interfaces run based on a  clock of 6MHz, which is unfortunately an exact sub-multiple of the codec reference.   So the difference between the clock in the transmitter (the codec) and the receiver (the PC) will cause a regular slippage.  

 

Now, soundcards communicate on USB using Isochronous mode which sacrifices data accuracy in favor of guaranteed timing of data delivery.   Other USB modes perform error checking and repeat failed packets, or are simple Human Interfaces (HID) like keyboards and mice which have low occasional data rate and no checking.

 

When you plug other devices into the bus, the negotiaion going on takes up some overhead and you'll lose more data blocks.

 

So, we are forced to accept data  glitches and resulting sampling rate errors in USB soundcards.

 

The only way to overcome this, is to force the USB to use another protocol, which means not using a plug-and-play soundcard chip / headphone dongle, or whatever.   This means a dedicated A/D converter, glue logic - which is probably going to have to be FPGA based, a PIC or like probably won't be able to hack-it - and a proper fully corrected USB link.  

 

Such interfaces already exist and are in the mainstearm of amateur radio. They are   Direct Sampling SDRs.  The SDR-IQ has a sampling rate defined only by its on board 66.6666MHz clock (which can come from a locked reference) and sends error corrected packets to the host.   Other radios use Ethernet instead of USB for a faster transfer rate.

 

That's life.   Mind you, when I get round to it, still going to get that chip operational with a locked reference 12MHz and accurately measure its sampling rate.   If it is not exactly 48kHz then we know the USB channel has to be at fault.

 

One thing that PCM2900 codec chip is useful for - its the only soundcard I know of that DOESN'T  allow you to change the recording level.   Voltage in is related to digits out - exactly.    So that makes it of inestimably more use in soundcard based test equipment as you don't have to keep making a level calibration check every other measurement.   As far as I recall, it is even calibrated by design, so a 1VRMS sine (2.82V pk-pk) is full scale.

 

Andy

www.g4jnt.com

 

On 10 March 2011 01:28, James Moritz <james.moritz@btopenworld.com> wrote:

Dear Andy, LF Group,

Not sure if the first mail I sent on this topic ever got through - not to me, anyway...

I have been using a USB sound card for VLF reception with my lap-top PC, and, like all the PCs/sound cardsI have tried, it shows what appear to be transient changes in the sample rate from time to time. It uses an integrated codec/usb interface chip similar to the device Andy mentioned, with a 12MHz clock crystal. I measured the actual clock frequency, which is about +47ppm from nominal. With a suitable stable calibration source, and 48kHz sample rate, Spec Lab indicated a sample rate error of -87ppm. The SR compensation facility in Spec Lab makes the indicated frequency on the spectrogram correct to within a fraction of 1ppm. The sample rate calibrator reflects exactly the slow thermal drift of the sound card crystal, less than 1ppm, but always with the much larger and constant offset of about 130ppm in total from what you would expect. Changing the sample rate to 44.1kHz gave an indicated sample rate error of -115ppm. I found by accident that plugging a USB memory stick into another USB port on the laptop caused the sample rate to change as well - just plugging it in gave a brief glitch of several ppm, but opening the folders on the USB stick resulted in an additional -81ppm shift, which remained until the USB stick was unplugged again, when the SR returned to the previous value. The actual 12MHz clock frequency was not noticeably affected by any of this activity.

In addition to these apparently stable shifts in sample rate, transients occur now and again of several ppm, aparently at random, maybe once an hour on average. These give rise to glitches in a strong trace on a spectrogram with resolution about 1mHz. But on weak amateur signals these would probably not be visible, since the transient sidebands mostly only last for a few pixels and are 20 or 30dB below the main spectral line, which remains on the correct frequency. However, I think when one gets into the microhertz resolution range, it would be a problem, because the transients occur often enough to merge together and smear out the spectral line.

Cheers, Jim Moritz
73 de M0BMU