Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-03 Thread glen english
ahh yes, my bad.
Alex, thanks for drawing that to my attention.


On 3/09/2015 10:03 PM, Alexandru Csete wrote:
> On Thu, Sep 3, 2015 at 3:38 AM, glen english  wrote:
>> Nice work Alex
>>
>> For a lump of code to be using THAT much CPU, and these CPUs are very
>> capable something must be terribly wrong. maybe turn some of the
>> optimizations off and see what happens.  Can you break it down to
>> computation effort (something stuck in a loop, loop variable getting
>> trashed leading to excessive iterations)  or  cache thrashing  ?
>>
> Hi Glen,
>
> Just to be clear, the numbers I posted do not express CPU load like
> e.g. top or time does. Instead they show how much % of the application
> execution time is spent in various functions. That's why I didn't post
> any hardware details.
>
> The perf tool is developed by linux kernel devs and is available on
> any linux system with kernel 2.6 or later:
> https://perf.wiki.kernel.org/index.php/Main_Page
>
> It can do much more advanced profiling than what I posted (also at
> assembly level), but even in its simplest form it gives a good idea of
> where an application spends most time.
>
> Alex
>
> --
> Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
> Get real-time metrics from all of your servers, apps and tools
> in one place.
> SourceForge users - Click here to start your Free Trial of Datadog now!
> http://pubads.g.doubleclick.net/gampad/clk?id=241902991


--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-03 Thread Alexandru Csete
On Thu, Sep 3, 2015 at 3:38 AM, glen english  wrote:
> Nice work Alex
>
> For a lump of code to be using THAT much CPU, and these CPUs are very
> capable something must be terribly wrong. maybe turn some of the
> optimizations off and see what happens.  Can you break it down to
> computation effort (something stuck in a loop, loop variable getting
> trashed leading to excessive iterations)  or  cache thrashing  ?
>

Hi Glen,

Just to be clear, the numbers I posted do not express CPU load like
e.g. top or time does. Instead they show how much % of the application
execution time is spent in various functions. That's why I didn't post
any hardware details.

The perf tool is developed by linux kernel devs and is available on
any linux system with kernel 2.6 or later:
https://perf.wiki.kernel.org/index.php/Main_Page

It can do much more advanced profiling than what I posted (also at
assembly level), but even in its simplest form it gives a good idea of
where an application spends most time.

Alex

--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-03 Thread wully
Hi

Some more analysis for the two modes:

Setup for freedv-rx in 700(b)-mode:

routineRXfreedv

input: 2.7kHz filtered samples @ 96kHz are coming in blocks of 1024 
Samples (cohpsk-data)

using libsamplerate downsampling from 48kHz to 7.5 kHz gives 80 
[email protected]
stored in input roundbuffer.
if ( more than 600 samples are avaliable)
{
  call to freedv_rx to typically get 640 Samples audio and
 upsampling and filtering with a sinc-FIR (41 coefficients) to 
96kHz to get 3840 Samples.
 stored in a output roundbuffer.
}
output: 1024 Samples@96kHz audiodata
(when output roundbuffer does not have enough samples, return an zero 
filled buffer).

elapsed time when not enough samples for freedv_floatrx: about 150 usec.
when freedv_floatrx is called, the elapsed time is about 4000 usec. 
(which occurs average every eights buffer).

total time in routineRXfreedv:
Anzahl verarbeitete Buffer: k=4072 (number of calls to the routine)
mode700B totalexecutiontime=2765887,demodcalls=542, 
longexecutiontime=2236199

so, about 0.5 sec are used only for downsampling etc.,



analysis with 1600-mode with the same input (about 43 seconds of audiodata):

routine has the same structure, but downsampling for fdmdv to 8kHz uses 
4 times 1024Samples@96kHz to get 320 Samples@8kHz fdmdv. (two calls to 
freedv_floatrx for fdmdv).
The downsampling is simple, because we have just to use every 12th 
sample (2.7kHz filtering done before the routineRXfreedv).

elapsed time without call to freedv_floatrx:about 12 usec.
elapsed time including call: less than 1800 usec (which occurs average 
about every fourth buffer).

Anzahl verarbeitete Buffer: k=4072
mode1600 totalexecutiontime=1864957,demodcalls=1087, 
longexecutiontime=1831541

so, most cycles go into the freedv_floatrx.

My conclusion: the libsamplerate (Secret Rabbit Code 
http://www.mega-nerd.com/SRC/) uses quite some cycles. But the main 
cycles are used by freedv_floatrx.

Alfred, hb9epu

P.S. I am using a Shuttle DS47 passive cooled tiny computer (20x4x16cm3) 
with a twocore intel cpu.



--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread David Rowe
Testing the core signal processing of the rx side:

~/codec2-dev/build_linux/src$ time ./freedv_rx 700 ~/Desktop/fade700.raw 
/dev/null

real0m0.799s
user0m0.795s
sys 0m0.004s

fade700 is a 60 second file, so that's 0.799/60 = 1.3% CPU.  It appears 
to be using just one core of my 4 core laptop (a modest, 4 year old 
machine).  700B won't be using any extra MIPs on rx.

That's with vanilla C code.  The FreeDV GUI program will be using 
somewhat more, as it's pushing pixels all over the screen.

Cheers,

David

On 03/09/15 11:08, glen english wrote:
> Nice work Alex
>
> For a lump of code to be using THAT much CPU, and these CPUs are very
> capable something must be terribly wrong. maybe turn some of the
> optimizations off and see what happens.  Can you break it down to
> computation effort (something stuck in a loop, loop variable getting
> trashed leading to excessive iterations)  or  cache thrashing  ?
>
>
>
> On 3/09/2015 9:04 AM, Alexandru Csete wrote:
>> On Wed, Sep 2, 2015 at 11:04 PM, David Rowe  wrote:
>>> Would some one on the list like to profile freedv_tx and freedv_rx and
>>> see where the MIPs are going?  That would be very helpful.
>>>
>
>
> --
> Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
> Get real-time metrics from all of your servers, apps and tools
> in one place.
> SourceForge users - Click here to start your Free Trial of Datadog now!
> http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
> ___
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>

--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread glen english
Nice work Alex

For a lump of code to be using THAT much CPU, and these CPUs are very 
capable something must be terribly wrong. maybe turn some of the 
optimizations off and see what happens.  Can you break it down to 
computation effort (something stuck in a loop, loop variable getting 
trashed leading to excessive iterations)  or  cache thrashing  ?



On 3/09/2015 9:04 AM, Alexandru Csete wrote:
> On Wed, Sep 2, 2015 at 11:04 PM, David Rowe  wrote:
>> Would some one on the list like to profile freedv_tx and freedv_rx and
>> see where the MIPs are going?  That would be very helpful.
>>


--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread Alexandru Csete
On Thu, Sep 3, 2015 at 1:18 AM, David Rowe  wrote:
> Thanks Alex,
>
> Yes that's interesting.  Looks like the rx is dominated by the receive
> filtering and for some reason there is power function being called quite
> a lot.  The powf can probably be moved outside of whatever loop it is in.
>
> H, not sure what kf_work does, I don't think it's one of my functions.

kf_work is the internal work function in kiss fft.

Alex

--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread David Rowe
Thanks Alex,

Yes that's interesting.  Looks like the rx is dominated by the receive 
filtering and for some reason there is power function being called quite 
a lot.  The powf can probably be moved outside of whatever loop it is in.

H, not sure what kf_work does, I don't think it's one of my functions.

Cheers,

David

On 03/09/15 08:34, Alexandru Csete wrote:
> On Wed, Sep 2, 2015 at 11:04 PM, David Rowe  wrote:
>>
>> Would some one on the list like to profile freedv_tx and freedv_rx and
>> see where the MIPs are going?  That would be very helpful.
>>
>
> Hi David,
>
> I have attached two text files containing the output of the perf tool,
> showing the relative distribution of CPU cycles, i.e. the sum of all
> is 100% though the list is incomplete. Was it something like this you
> had in mind?
>
> I captured it using:
>  perf record 
>
> then show results:
>  perf report
>
> Alex
>
>
>
> --
> Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
> Get real-time metrics from all of your servers, apps and tools
> in one place.
> SourceForge users - Click here to start your Free Trial of Datadog now!
> http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
>
>
>
> ___
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>

--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread Alexandru Csete
On Wed, Sep 2, 2015 at 11:04 PM, David Rowe  wrote:
>
> Would some one on the list like to profile freedv_tx and freedv_rx and
> see where the MIPs are going?  That would be very helpful.
>

Hi David,

I have attached two text files containing the output of the perf tool,
showing the relative distribution of CPU cycles, i.e. the sum of all
is 100% though the list is incomplete. Was it something like this you
had in mind?

I captured it using:
perf record 

then show results:
perf report

Alex
 40.20%  freedv_rx  libcodec2.so.0.4   [.] rx_filter_coh
 21.59%  freedv_rx  libm-2.19.so   [.] __powf_finite
 17.20%  freedv_rx  libcodec2.so.0.4   [.] kf_work
  4.34%  freedv_rx  libcodec2.so.0.4   [.] fdm_downconvert_coh
  2.58%  freedv_rx  libm-2.19.so   [.] __sincosf
  2.26%  freedv_rx  libcodec2.so.0.4   [.] aks_to_M2
  2.11%  freedv_rx  libcodec2.so.0.4   [.] lpc_post_filter
  2.06%  freedv_rx  libm-2.19.so   [.] __atanf
  1.14%  freedv_rx  libcodec2.so.0.4   [.] synthesise
  0.83%  freedv_rx  libcodec2.so.0.4   [.] rx_est_timing
  0.80%  freedv_rx  libm-2.19.so   [.] __powf
  0.72%  freedv_rx  libcodec2.so.0.4   [.] phase_synth_zero_order
  0.59%  freedv_rx  [kernel.kallsyms]  [k] 0x8104f45a
  0.59%  freedv_rx  libm-2.19.so   [.] __atan2f_finite
  0.39%  freedv_rx  libcodec2.so.0.4   [.] postfilter
  0.37%  freedv_rx  libm-2.19.so   [.] __ieee754_atan2_fma4
  0.31%  freedv_rx  libcodec2.so.0.4   [.] fdmdv_freq_shift_coh
  0.24%  freedv_rx  libcodec2.so.0.4   [.] corr_with_pilots
  0.21%  freedv_rx  libcodec2.so.0.4   [.] qpsk_symbols_to_bits
  0.16%  freedv_rx  libcodec2.so.0.4   [.] cohpsk_demod
  0.16%  freedv_rx  libcodec2.so.0.4   [.] synthesise_one_frame
  0.11%  freedv_rx  libcodec2.so.0.4   [.] freedv_rx
  0.10%  freedv_rx  libcodec2.so.0.4   [.] lsp_to_lpc
  0.09%  freedv_rx  libm-2.19.so   [.] __atan2f
  0.07%  freedv_rx  libm-2.19.so   [.] __logf_finite
  0.07%  freedv_rx  libm-2.19.so   [.] __cosf
  0.05%  freedv_rx  libm-2.19.so   [.] __ieee754_pow_fma4
  0.05%  freedv_rx  libcodec2.so.0.4   [.] codec2_decode_700b
  0.05%  freedv_rx  libcodec2.so.0.4   [.] linreg
  0.05%  freedv_rx  libcodec2.so.0.4   [.] rate_Fs_rx_processing
  0.04%  freedv_rx  libcodec2.so.0.4   [.] update_ct_symb_buf
  0.04%  freedv_rx  libc-2.19.so   [.] vfprintf
  0.04%  freedv_rx  libcodec2.so.0.4   [.] freedv_comprx
  0.04%  freedv_rx  libc-2.19.so   [.] __printf_fp
  0.04%  freedv_rx  libcodec2.so.0.4   [.] codec2_rand
  0.03%  freedv_rx  libm-2.19.so   [.] __exp1_fma4
  0.03%  freedv_rx  libc-2.19.so   [.] __memcpy_sse2
P 23.68%  freedv_tx  libcodec2.so.0.4   [.] tx_filter_and_upconvert_coh
 23.28%  freedv_tx  libcodec2.so.0.4   [.] kf_work
 16.07%  freedv_tx  libcodec2.so.0.4   [.] inverse_filter
 15.79%  freedv_tx  libcodec2.so.0.4   [.] nlp
  5.67%  freedv_tx  libcodec2.so.0.4   [.] hs_pitch_refinement
  3.98%  freedv_tx  libcodec2.so.0.4   [.] mbest_search
  2.70%  freedv_tx  libcodec2.so.0.4   [.] est_voicing_mbe
  2.14%  freedv_tx  libcodec2.so.0.4   [.] estimate_amplitudes
  1.12%  freedv_tx  libcodec2.so.0.4   [.] autocorrelate
  0.75%  freedv_tx  libcodec2.so.0.4   [.] cheb_poly_eva
  0.75%  freedv_tx  libcodec2.so.0.4   [.] dft_speech
  0.64%  freedv_tx  [kernel.kallsyms]  [k] 0x8104f45a
  0.54%  freedv_tx  libcodec2.so.0.4   [.] lpc_to_lsp
  0.52%  freedv_tx  libcodec2.so.0.4   [.] cohpsk_clip
  0.46%  freedv_tx  libcodec2.so.0.4   [.] analyse_one_frame
  0.25%  freedv_tx  libcodec2.so.0.4   [.] speech_to_uq_lsps
  0.22%  freedv_tx  libcodec2.so.0.4   [.] codec2_encode_700b
  0.22%  freedv_tx  libm-2.19.so   [.] __powf_finite
  0.22%  freedv_tx  libcodec2.so.0.4   [.] post_process_sub_multiples
  0.12%  freedv_tx  libcodec2.so.0.4   [.] freedv_comptx
  0.12%  freedv_tx  libm-2.19.so   [.] __ieee754_log_fma4
  0.09%  freedv_tx  libm-2.19.so   [.] __logf_finite
  0.07%  freedv_tx  libcodec2.so.0.4   [.] freedv_tx
  0.06%  freedv_tx  libcodec2.so.0.4   [.] levinson_durbin
  0.06%  freedv_tx  libm-2.19.so   [.] __log10f_finite
  0.06%  freedv_tx  libcodec2.so.0.4   [.] two_stage_pitch_refinement
  0.05%  freedv_tx  libcodec2.so.0.4   [.] lspmelvq_mbest_encode
  0.04%  freedv_tx  libc-2.19.so   [.] _int_free
  0.04%  freedv_tx  libc-2.19.so   [.] __GI___mempcpy
  0.04%  freedv_tx  libc-2.19.so   [.] malloc
  0.04%  freedv_tx  libm-2.19.so   [.] __log10_finite
  0.02%  freedv_tx  libm-2.19.so   [.] __acosf_finite
  0.02%  freedv_tx  libcodec2.so.0.4   [.] cohpsk_mod
  0.02%  freedv_tx  libcodec2.so.0.4   [.] mbest_create
  0.02%  freedv_tx  libc-2.19.so   [.] _int_malloc
  0.02%  freedv_tx  libcodec2.so.0.4   [.] ceilf@plt
  0.01%  freedv_tx  libcodec2.so.0.4   [.] bits_to_qpsk_symbols
--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place

Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread David Rowe
IIRC Alfred has a custom SDR, so the connection between his radio and 
the FreeDV API is software - floating point samples.  This is an 
interesting interface and may be related to the inaccurate SNR measures 
he is getting.

It is also an interesting pointer to the future.  FreeDV can be fed 
samples directly from the ADC or digital down converter. No "gain" or 
AGC is required. If you feed the FreeDV API samples scaled by 0.0001 or 
1 it will work exactly the same.

Much of the signal processing chain we use for SSB is going to evaporate 
in the next few years.

Cheers,

David

On 03/09/15 05:54, Walter Holmes wrote:
> Alfred,
>
>   Just for a point of reference, how are you connected to your radio?
>
> Ie.. are you USB, LAN cable, or via a digital interface of some kind?
>
> I am using an HPSDR radio attaching to my radio via a LAN cable, and I have
> heard from others as well that are USB attached that they saw an increase in
> CPU in the 700/700b modes.
>
> But I do not see that increase at all in my environment.
>
> Thanks,
>
> Walter/K5WH
>
> -Original Message-
> From: wully [mailto:[email protected]]
> Sent: Wednesday, September 02, 2015 1:59 PM
> To: [email protected]
> Subject: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx
> than the 1600mode
>
> Hi
>
> I am experimenting further with the 700B mode. It seems that this mode uses
> quite a bit of cpu cycles during rx, substantially more than the
> 1600 mode. I am using a relatively slow cpu, which runs my sdr code at about
> 20-25% cpu in SSB, when I switch to 700B rx, the cpu increases to about
> 45-50%.
>
> I assume, that the 700B makes extensive calculations?
>
>
> Question about SNR-estimation:
>
> I just display the value codec2snr that I get from calling
>
> int codec2sync;
> float codec2snr;
>
> freedv_get_modem_stats(freedvRx,&codec2sync,&codec2snr);
>
> What are the typical values that should appear?
>
> Thanks
> Alfred, hb9epu
>
>
>
> 
> --
> Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
> Get real-time metrics from all of your servers, apps and tools in one place.
> SourceForge users - Click here to start your Free Trial of Datadog now!
> http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
> ___
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>
>
> --
> Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
> Get real-time metrics from all of your servers, apps and tools
> in one place.
> SourceForge users - Click here to start your Free Trial of Datadog now!
> http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
> ___
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>

--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread David Rowe
Hi Alfred,

I haven't actually measured the CPU load of 700 yet.  Often it's just 
one small function that is taking all of the MIPs.  The current CPU load 
of 700B will have no relationship to the CPU load when optimised.

Would some one on the list like to profile freedv_tx and freedv_rx and 
see where the MIPs are going?  That would be very helpful.

The SNR values should be roughly the SNR of the channel, so at the lower 
limit (when 700B is breaking down) about 0dB.  In a fade it may drop to 
negative values. It reads correctly for me.

The cohpsk_demod program can output SNR for you - maybe try running this 
program with a file of Fs=7500Hz, 16 bit samples from your off-air signal.

You could also try running the FreeDV GUI program with off air samples 
or connected to another receiver.  See what SNR values it presents.

It may be some issue with the SNR routine that you have found, in that 
case we can track it down with some more tests.

Cheers,

David

On 03/09/15 04:28, wully wrote:
> Hi
>
> I am experimenting further with the 700B mode. It seems that this mode
> uses quite a bit of cpu cycles during rx, substantially more than the
> 1600 mode. I am using a relatively slow cpu, which runs my sdr code at
> about 20-25% cpu in SSB, when I switch to 700B rx, the cpu increases to
> about 45-50%.
>
> I assume, that the 700B makes extensive calculations?
>
>
> Question about SNR-estimation:
>
> I just display the value codec2snr that I get from calling
>
> int codec2sync;
> float codec2snr;
>
> freedv_get_modem_stats(freedvRx,&codec2sync,&codec2snr);
>
> What are the typical values that should appear?
>
> Thanks
> Alfred, hb9epu
>
>
>
> --
> Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
> Get real-time metrics from all of your servers, apps and tools
> in one place.
> SourceForge users - Click here to start your Free Trial of Datadog now!
> http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
> ___
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>

--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


Re: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx than the 1600mode

2015-09-02 Thread Walter Holmes
Alfred,

 Just for a point of reference, how are you connected to your radio?

Ie.. are you USB, LAN cable, or via a digital interface of some kind?

I am using an HPSDR radio attaching to my radio via a LAN cable, and I have
heard from others as well that are USB attached that they saw an increase in
CPU in the 700/700b modes.

But I do not see that increase at all in my environment.

Thanks,

Walter/K5WH

-Original Message-
From: wully [mailto:[email protected]] 
Sent: Wednesday, September 02, 2015 1:59 PM
To: [email protected]
Subject: [Freetel-codec2] 700B mode seems to use much more cpu cycles in rx
than the 1600mode

Hi

I am experimenting further with the 700B mode. It seems that this mode uses
quite a bit of cpu cycles during rx, substantially more than the
1600 mode. I am using a relatively slow cpu, which runs my sdr code at about
20-25% cpu in SSB, when I switch to 700B rx, the cpu increases to about
45-50%.

I assume, that the 700B makes extensive calculations?


Question about SNR-estimation:

I just display the value codec2snr that I get from calling

int codec2sync;
float codec2snr;

freedv_get_modem_stats(freedvRx,&codec2sync,&codec2snr);

What are the typical values that should appear?

Thanks
Alfred, hb9epu




--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


--
Monitor Your Dynamic Infrastructure at Any Scale With Datadog!
Get real-time metrics from all of your servers, apps and tools
in one place.
SourceForge users - Click here to start your Free Trial of Datadog now!
http://pubads.g.doubleclick.net/gampad/clk?id=241902991&iu=/4140
___
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2