Occasionally false samples from a soundcard
 

A PCI bus based "professional" soundcard has a digital input (S/PDIF).
Connected this input to a CD player's output and recorded about 25 seconds
using waverecorder.

Found 14 "clicks" in the recorded WAV file.
These clicks are false sample values, *exactly one* false sample in each
case
Sample value is random (never full scale nor zero).

Searched for the clicks and determined their sample number.
12 of them have a sample number ending with "1", 2 of them end with "2", so
definitely *not random* (you may detect other non-randomness in these
numbers as well, see details below).

Calculated the difference between sample numbers.
Calculated the gcd (greatest common divisor) of these differences and found
that it is always 440 (or multiples of 440, e.g. 880, 2200).

My questions:
Do you have any idea what goes wrong here?
What may the 440 stand for?

Juergen


The details:

Soundcard is based on an ICE-1712 Envy24 PCI interface.
SPDIF interface is Cirrus CS8427.
A/D-D/A is AKM AK4524.
OS is WinXP Prof. SP2.
Mainboard is ASUS P4PE, chipset is Intel 845PE (82801).
CPU is Pentium 4, 2.8 GHz, RAM is 1GB.
Soundcard sits in PCI slot 5, IRQ 9, slot 5 shares IRQ only with (empty)
slot 1, so IRQ 9 is exclusively for soundcard.

Recording application is WaveLab Lite 2.0 (but getting comparable results
with Audacity, Windows recorder...).

Getting clicks no matter if using digital/SPDIF or analog inputs.
Getting same type of clicks (exactly one false sample) no matter of what
sample rate using (tried 22.050k to
96.000k).
Getting clicks only if some "level" is there, getting no clicks if recording
silence.
Absolute no clicks are heard on line out/headphone while recording.

These are the sample numbers of false samples for SPDIF optical in at 44.1k
(always exactly one false sample) as shown by Audacity:

0 start of silence
177074 (ca.) start of music
182602 4,140635 sec
194481 4,410000 sec diff: 11879
247281 5,607279 sec diff: 52800 (gcd : 1320; for 11880)
248161 5,627234 sec diff: 880 gcd : 880
279841 6,345601 sec diff: 31680 gcd : 880
459801 10,426327 sec diff:179960 gcd : 440
492801 11,174626 sec diff: 33000 gcd : 440
523161 11,863061 sec diff: 30360 gcd : 1320
705761 16,003651 sec diff:182600 gcd : 440
726001 16,462608 sec diff: 20240 gcd : 440
754602 17,111156 sec diff: 28601 (gcd : 440; for 28600)
796401 18,058980 sec diff: 41799 (gcd : 2200, for 41800/28600)
999241 22,658526 sec diff:202840 (gcd : 440; for 41800)
1049841 23,805918 sec diff: 50600 gcd : 440

Wondering if these PCI Config Space entries are ok:
Soundcard is a PCI busmaster, but has "Min_Gnt = 0" and "Max_Lat = 0".

Juergen

Occasionally false samples from a soundcard
 

In ,
Juergen Marquardt typed:

Calculated the difference between sample numbers.
Calculated the gcd (greatest common divisor) of these differences and
found that it is always 440 (or multiples of 440, e.g. 880, 2200).

My questions:
Do you have any idea what goes wrong here?
What may the 440 stand for?


Seems to me that the SB driver software is buffering the data before writing
to the file and the buffer size is 440 bytes and there is a bug in the
driver.

My guess.

Jo

Occasionally false samples from a soundcard
 

Juergen Marquardt wrote:
A PCI bus based "professional" soundcard has a digital input (S/PDIF).
Connected this input to a CD player's output and recorded about 25 seconds
using waverecorder.

Found 14 "clicks" in the recorded WAV file.
These clicks are false sample values, *exactly one* false sample in each
case
Sample value is random (never full scale nor zero).

Searched for the clicks and determined their sample number.
12 of them have a sample number ending with "1", 2 of them end with "2", so
definitely *not random* (you may detect other non-randomness in these
numbers as well, see details below).

Calculated the difference between sample numbers.
Calculated the gcd (greatest common divisor) of these differences and found
that it is always 440 (or multiples of 440, e.g. 880, 2200).

My questions:
Do you have any idea what goes wrong here?
What may the 440 stand for?

Juergen


The details:

Soundcard is based on an ICE-1712 Envy24 PCI interface.
SPDIF interface is Cirrus CS8427.
A/D-D/A is AKM AK4524.
OS is WinXP Prof. SP2.
Mainboard is ASUS P4PE, chipset is Intel 845PE (82801).
CPU is Pentium 4, 2.8 GHz, RAM is 1GB.
Soundcard sits in PCI slot 5, IRQ 9, slot 5 shares IRQ only with (empty)
slot 1, so IRQ 9 is exclusively for soundcard.

Recording application is WaveLab Lite 2.0 (but getting comparable results
with Audacity, Windows recorder...).

Getting clicks no matter if using digital/SPDIF or analog inputs.
Getting same type of clicks (exactly one false sample) no matter of what
sample rate using (tried 22.050k to
96.000k).
Getting clicks only if some "level" is there, getting no clicks if recording
silence.
Absolute no clicks are heard on line out/headphone while recording.

These are the sample numbers of false samples for SPDIF optical in at 44.1k
(always exactly one false sample) as shown by Audacity:

0 start of silence
177074 (ca.) start of music
182602 4,140635 sec
194481 4,410000 sec diff: 11879
247281 5,607279 sec diff: 52800 (gcd : 1320; for 11880)
248161 5,627234 sec diff: 880 gcd : 880
279841 6,345601 sec diff: 31680 gcd : 880
459801 10,426327 sec diff:179960 gcd : 440
492801 11,174626 sec diff: 33000 gcd : 440
523161 11,863061 sec diff: 30360 gcd : 1320
705761 16,003651 sec diff:182600 gcd : 440
726001 16,462608 sec diff: 20240 gcd : 440
754602 17,111156 sec diff: 28601 (gcd : 440; for 28600)
796401 18,058980 sec diff: 41799 (gcd : 2200, for 41800/28600)
999241 22,658526 sec diff:202840 (gcd : 440; for 41800)
1049841 23,805918 sec diff: 50600 gcd : 440

Wondering if these PCI Config Space entries are ok:
Soundcard is a PCI busmaster, but has "Min_Gnt = 0" and "Max_Lat = 0".

Juergen




My best guess, from experience with Creative cards, would be drivers -
perhaps have a look around for some beta drivers? Also, I'd have a
fiddle with the latency setting - zero seems a little infeasible?! Try
40-50ms.

Rob

Occasionally false samples from a soundcard
 

In ,
Jo typed:
In ,
Juergen Marquardt typed:

Calculated the difference between sample numbers.
Calculated the gcd (greatest common divisor) of these differences and
found that it is always 440 (or multiples of 440, e.g. 880, 2200).

My questions:
Do you have any idea what goes wrong here?
What may the 440 stand for?


Seems to me that the SB driver software is buffering the data before
writing to the file and the buffer size is 440 bytes and there is a
bug in the driver.

My guess.

Jo

....or buffer size is 440 DWORDS

Jo