Hi Jan,
> Thanks. I continue to play with different settings for sound card and fount
> out some dependencies. When I set DMA rate bellow 93, I get sometimes only
> one RX overrun_error and RX A/D speed error with speed bellow 48000 - from
> about 40000 to 42500.
Presumably you would see an ugly spectrum if you would send a good
sinewave into the computer under these conditions. I would guess
that the soundcard runs at 48kHz but that sampling is not
continous. Each time a discontinuity occurs there would be
a wideband click that would easily be seen on the spectrum
and on the waterfall.
> When I set DMA rate 93 and above then there in no RX
> overrun_error and the RX A/D speed above 48000 - about 49000.
There should be nothing wrong in not limiting the DMA rate.
The bandwidth you set will determine the buffer size for the
soundcard. A narrow bandwidth for fft1 or fft2 may cause the
same problem with the soundcard as a DMA rate limit.
The reported speed of 49 kHz seems odd. What happens when you
leave Linrad running for 5 minutes with 'T' pressed to display
the sampling rate? For this test, do not click any signal, just
monitor the input rate.
> Only Linrad and anti-virus program is running, Linrad reports load bellow
> 2%, overall CPU load is bellow 3 %. I will continue with investigation and
> report any improvements.
The problem can not be CPU overload. Something is wrong because the
operating system should give an error message if the device driver
can not use the larger buffers associated with lower DMA rates.
> Do you think that step down to Windows XP could help?
I have no idea. I have just recently become aware of this class
of problems because I have them on my new hardware. I have a feeling
that the problem is not the soundcard itself, but some other drivers
that are needed for USB, PCI or whatever.
> I am decided to have
> separate PC for Linrad. I understood from your comment that single core
> could be better. Can anyone recommend some PC configuration that works well?
I have never seen any problem of this kind in the Pentium II to Pentium IV
computers. They are old now, but for running at 48 kHz anything you can
find in the junk-yard should be fine:-)
I have not much experience with modern computers with modern operating
systems, but I have recently started to investigate what works (not much)
and what does not work (nearly everything) on my D5400XS Intel motherboard
with two quad core Xeon processors. Intel specifies that it should
work with both 32 and 64 bit versions of XP and Vista, but nothing else.
That is not quite correct, it works perfectly well with reasonably modern
Linux and also with Windows 7 but I can not find correctly working
drive routines for most of my (old) soundcards. It would have been
interesting to see how Windows 2000 would behave, but I have not been
able to install it on the D5400XS.
The problems I see with Delta 44 and Soundbalster Audigy are present
also when I run Winrad so the problems are not related to Linrad.
The Vista drivers seem to work with Vista on older computers having
a single CPU. I do not know whether the problems are because of the
PCI bus or whether it is due to multiple processors (or something
else.) Linrad works perfectly well under all operating systems that
I have been able to install when input is a file on the hard disk.
It is interesting to note that the CPU load for individual threads
differs significantly. The reason probably being poor usage of cache
in older operating systems.
Right now I am in the process of documenting the timing tests when
running Linrad-03.07 with very high CPU load from a 2 MHz recording.
With parameters that make the fft1 thread run at about 90% CPU usage
on most of the operating systems, the oldest Linux, Ubuntu 6.10
needs a little more than 100% so it has short gaps in the output
(overrun errors) while 32 bit XP uses 98% with no errors.
The 64 bit version of XP uses 92% and so does both versions of Vista
while the 64 bit version of Windows 7 uses 75%. The most recent
version of Linux uses 88% (64 bit Fedora 11).
I do not know whether the CPU load per thread that is reported to
Linrad is accurate under all operating systems. I will increase
the CPU load until processing fails to check. Under Ubuntu 6.10
it seems OK, but under Debian Lenny the timings for individual
threads seems to be incorrect.
Investigating (and documenting) the problems with the soundcards
will be the next project once 03-07 and associated timing tests are
finalized.