Commit Graph

665 Commits

Author SHA1 Message Date
Jared Boone
d9bfaaf805 Stop baseband streaming when switching radio configuration.
Probably unnecessary, but feeling paranoid about changing sample rates while there's a processor actively handling samples, and potentially maxing out the M4 core.
2016-01-10 11:01:54 -08:00
Jared Boone
136ba895ef Clear spectrum FIFO when streaming is stopped. 2016-01-10 11:00:09 -08:00
Jared Boone
62e5ec1dbb Postpone clearing channel_spectrum_request_update until finished with data in array.
There was potential for new samples to be copied into array as the FFT and post-processing were occuring.
2016-01-10 10:45:22 -08:00
Jared Boone
d821afc60d Spectrum streaming control, spectrum attributes back in each frame.
TODO: This feels kinda complex, and there's some repeated Processor code that needs to be refactored into a base class.
2016-01-10 10:42:20 -08:00
Jared Boone
0647f26707 Move SGPIO configuration and control to baseband firmware.
Addresses long-standing and annoying bug where SGPIO DMA channel would not disable -- and not configure cleanly next time it was needed. My theory is that the DMA channel couldn't disable until it got a request from the peripheral, and sometimes the peripheral was disabled before that last request.

Anyway, the baseband firmware should control the SGPIO, methinks, despite the impact on baseband code size.
2016-01-10 10:23:39 -08:00
Jared Boone
8fde4972b4 Methods to peek and skip messages in queue.
Allows receiver to not consume a message until after it's handled. And that enables the transmitter to block until the queue is empty, knowing that when unblocked, all messages in queue have been handled.
2016-01-10 10:15:25 -08:00
Jared Boone
c75c167c25 Rename GPDMA channel disable_force() to disable().
There was nothing particularly forceful about the disable_force() code. Whether it's a "forced" operation depends on what happens before and after (HALT and ACTIVE flags are involved in a smooth disabling of the channel).
2016-01-10 10:11:16 -08:00
Jared Boone
43d6098c48 Move baseband_sgpio.* to common/.
It'll soon be used by M4 code, not M0. SGPIO baseband streaming needs to be tightly controlled by... the baseband!
2016-01-09 17:57:37 -08:00
Jared Boone
365c2ef946 Handle baseband::dma::wait_for_rx_buffer() returning empty buffer.
Was technically OK before, because sample count was zero. But seems silly (and vaguely dangerous) to call all that code with a nullptr.
2016-01-09 12:20:57 -08:00
Jared Boone
c3167ac27c Missing #includes. 2016-01-09 12:17:17 -08:00
Jared Boone
8fba47b7d8 FIFO in and out indices should be volatile.
Later code revealed that the receiving core never notices (reads anew) the _in member variable when waiting for the FIFO to be empty (_in == _out).
2016-01-07 10:53:27 -08:00
Jared Boone
7710b2d1fa ChannelSpectrumConfig message subsumes FIFONotify.
Separate channel spectrum config from spectrum data. This will permit sending config info only when necessary.
Use type information of ChannelSpectrum to statically define number of FFT bins elsewhere.
TODO: Posting configuration message way too often. Fixing that is the next step.
2016-01-06 12:10:30 -08:00
Jared Boone
ba33cc737d Transmit DisplayFrameSync message, handle in WaterfallSpectrum.
More kludginess, especially around initialization and timing. But it addresses the flickering lines of pixels at the bottom of the waterfall scroll area!
2016-01-06 11:36:57 -08:00
Jared Boone
aaa1bc3a09 Transmit ChannelSpectrum data through separate FIFO.
Allows handling of data during LCD "vertical retrace", independent of other baseband->application messages. A bit kludgy still...
2016-01-06 11:34:41 -08:00
Jared Boone
2cd28fcc0c Make FIFO::in() more consistent with other functions. 2016-01-06 11:04:25 -08:00
Jared Boone
4059e9cebe Reintroduce FIFO in/out for templated FIFO element type. 2016-01-06 11:01:02 -08:00
Jared Boone
b05923eab6 Change FIFO is_full() to make more sense. 2016-01-06 10:58:42 -08:00
Jared Boone
ea8c186fcf Revert "Try to fix dumb S3 URLs."
This reverts commit 794925442e.
2016-01-05 21:29:40 -08:00
Jared Boone
794925442e Try to fix dumb S3 URLs. 2016-01-05 21:04:35 -08:00
Jared Boone
dcb2c46c57 Add wideband FM spectrum waterfall. 2016-01-04 17:27:18 -08:00
Jared Boone
36021689f9 Remove commented, dead code. 2016-01-04 15:42:19 -08:00
Jared Boone
73671ca5f6 Update sample rate and buffer size comments in execute() methods. 2016-01-04 15:41:49 -08:00
Jared Boone
c1e5577669 Eliminate duplicate code between event_m4.hpp, event.hpp.
event.* is common code, _m0 and _m4 contain specific event mask definitions.
2016-01-04 12:54:05 -08:00
Jared Boone
229c1e9a3c Move M4 MAPP IRQ handler to be with the event handler class.
event.[ch]pp and event_m4.[ch]pp are looking almost the same now...
2016-01-04 12:44:28 -08:00
Jared Boone
0fe5409357 Move M4TXEVENT interrupt control into event handler run(). 2016-01-04 12:15:35 -08:00
Jared Boone
69b258cc2b Consolidate event handler code, including IRQ. 2016-01-04 12:15:10 -08:00
Jared Boone
b058c0fe00 Consolidate M0-specific code. 2016-01-04 12:07:30 -08:00
Jared Boone
a27489eb21 Consolidating M4-specific saturation flag code. 2016-01-04 12:00:52 -08:00
Jared Boone
b42d3cc9d5 Move BasebandStatsCollector code into .cpp, as much as possible.
Still have template for callback, but hopefully will mitigate bloat when BasebandStatsCollector is used in multiple places.
2016-01-04 11:49:39 -08:00
Jared Boone
6d3cdb9bbf Separate template/callback code from method code.
...for more small reductions in code size.
2016-01-04 11:45:33 -08:00
Jared Boone
cd3a5afdb1 More templating elimination (MatchedFilter), some #include clean-up.
592 bytes!
2016-01-04 11:32:47 -08:00
Jared Boone
e6f69c90f2 Reduce inlining of FIR configuration due to templating.
80 bytes!
2016-01-04 11:20:28 -08:00
Jared Boone
359fb6f3c6 ...and more pass-by-reference.
Another 176 bytes!
2016-01-04 11:04:28 -08:00
Jared Boone
05a1c22426 More tweaks to pass buffers by reference.
144 bytes less code!
2016-01-04 11:00:17 -08:00
Jared Boone
778e2d2381 A little code consolidation -- M0APP IRQ. 2016-01-04 10:12:18 -08:00
Jared Boone
cd17f414b7 ChannelDecimator now unused, remove from Makefile, remove commented code. 2016-01-04 09:49:48 -08:00
Jared Boone
168bd02dea Move AudioStatsCollector code into .cpp. 2016-01-04 09:25:43 -08:00
Jared Boone
c5b234d9f6 Remove BasebandProcessor virtual method replaced by message. 2016-01-04 09:13:17 -08:00
Jared Boone
222b878629 Use BasebandProcessor::mute_audio(), remove I2S #include dependencies. 2016-01-04 09:07:07 -08:00
Jared Boone
ef6ae8ed94 Add mute_audio() function to baseband processor. 2016-01-04 09:06:17 -08:00
Jared Boone
8fb3ffaedc Reorder squelch code to group things a bit more logically. 2016-01-04 09:01:51 -08:00
Jared Boone
b0855b3c4d Pass buffers by reference to save significant text section code size.
752 bytes!
2016-01-03 23:09:06 -08:00
Jared Boone
b03da8d693 Fix calculation of work_audio_buffer item count. 2016-01-03 22:42:40 -08:00
Jared Boone
d9017530c6 Moved baseband temporary buffers back into class member variables.
Significant performance hit when declaring on stack. C++ wants to initialize std::array or even raw array with element constructors, was resulting in ~7% CPU utilization, for a buffer that was immediately written over anyway.
2016-01-03 22:31:44 -08:00
Jared Boone
5547782f5a Improve FFT twiddle factor precisions using Python SymPy.
Python math, numpy, scipy produce values with significant rounding errors.
2016-01-03 21:02:44 -08:00
Jared Boone
963b6e257a FFT twiddles can be used for all K <= table length.
Change assert to allow FFTs < 8.
2016-01-03 20:25:31 -08:00
Jared Boone
54e59cc224 Hard-code K=8 FFT twiddle constants for now.
So low on baseband RAM! This eliminates the sin_f32() function and table, which is only used in the FFT, for the moment.
2016-01-03 20:08:26 -08:00
Jared Boone
65dd385c94 Little FFT loop indexing tweak. 2016-01-03 17:47:08 -08:00
Jared Boone
78992c2f8d Adjust default FM squelch threshold. 2016-01-03 17:05:00 -08:00
Jared Boone
8adaddac5f Make FM squelch runtime-configurable. 2016-01-03 17:03:16 -08:00