VGAtoBaseband: Unterschied zwischen den Versionen

Aus LaborWiki
Wechseln zu: Navigation, Suche
K (+arguments description)
K (+more documentation)
Zeile 1: Zeile 1:
{{ProjektInfoBox
{{ProjektInfoBox
|name        = VGAtoIQBaseband
|name        = VGAtoIQBaseband
|status      = experimental
|status      = stable
|image      =  
|image      =  
|description = use the VGA port to generate I/Q baseband signals
|description = use the VGA port to generate I/Q baseband signals
Zeile 9: Zeile 9:
|update      =  
|update      =  
|platform    = Linux
|platform    = Linux
|license    = GPL
|license    = mixed
|download    = [https://www.das-labor.org/trac/browser/host-apps/VGAtoIQBaseband]
|download    = [https://www.das-labor.org/trac/browser/host-apps/VGAtoIQBaseband]
}} __NOTOC__
}} __NOTOC__
Zeile 62: Zeile 62:
== Arguments ==
== Arguments ==


  ./vgatoiqbaseband inputfile [ARGS]
  ./vgatoiqbaseband [inputfile] [ARGS]


if inputfile isn't specified, use -t argument
if no inputfile is specified, -t is used


ARGS could be:
ARGS could be:
Zeile 101: Zeile 101:
  ./vgatoiqbaseband -v -t
  ./vgatoiqbaseband -v -t
This will generate test paterns, pixelclk is 64Mhz, it will output 64/7 = 9.142 MSPS, set HSYNC to 1 (7 pixels) and VSYNC to 3 lines.<br>
This will generate test paterns, pixelclk is 64Mhz, it will output 64/7 = 9.142 MSPS, set HSYNC to 1 (7 pixels) and VSYNC to 3 lines.<br>
Will work on all GPUs.
Might work on all GPUs.
This also removes the right-most data value in each line and cuts off the 3 bottom-most lines from the input data to prevent phase shifts.
This also removes the right-most data value in each line and cuts off the 3 bottom-most lines from the input data to prevent phase shifts.


Zeile 122: Zeile 122:


{| {{Prettytable}}
{| {{Prettytable}}
! DAC clock [Mhz]!! Msymbols/s!! carrier ratio !! Bandwidth
! DAC clock [Mhz]!! Msymbols/s!! carrier ratio !! 3dB Bandwidth Mhz
|-
|-
|64
|64
Zeile 147: Zeile 147:
To remove those artefakts a si-filter is neccessary, including an interpolation DAC, which is running at much higher sampling rate.<br>
To remove those artefakts a si-filter is neccessary, including an interpolation DAC, which is running at much higher sampling rate.<br>
This is done by the fragment shader, implemented in shader.cpp.
This is done by the fragment shader, implemented in shader.cpp.
Modify KERNEL_SIZE defined in shader.cpp to set the convolutional depth.


*Note* The GPU has to fetch,multiply, add and store at least
The GPU has to fetch,multiply, add and store at least ''' <sampling rate> * <conv_depth> ''' pixels
<sampling rate> * KERNEL_SIZE pixels
 
only odd values can be used !


{| {{Prettytable}}
{| {{Prettytable}}
! DAC clock [Mhz]!! GPU !! Driver !! max Kernel_size
! DAC clock [MSps]!! GPU !! Driver !! max Kernel_size
|-
|-
|64
|64
Zeile 166: Zeile 162:
|fglrx 8.97.2
|fglrx 8.97.2
|17
|17
|-
|recommended
|
|
|27 or more
|}
|}



Version vom 22. März 2013, 18:26 Uhr

         
VGAtoIQBaseband

Release status: stable [box doku]

Description use the VGA port to generate I/Q baseband signals
Author(s)  siro (siro)
Last Version  0.1 ()
Platform  Linux
License  mixed
Download  [1]




Description

This software generates analog I/Q Signals that can be feed into a rf-modulator. This software is intended to be used with SDRT.

Requirements

  • PC with VGA adapter
  • OpenGL with Direct Redering
  • Vsync enabled
  • Software-defined Radio Transmitter that generates I/Q Signals
  • GL_ARB_fragment_program to use the low-pass filter
  • GL_ARB_pixel_buffer_object for efficient CPU to GPU transfers

What it does

Vgatoiqbaseband.png

Linux

tested with:

  • Mesa 9.0.2
  • fglrx 8.97.2

Windows

NOT SUPPORTED

  • it is possible to add custom edids / modelines, but that's rather a hack using modified inf files
  • other ways of adding modelines are unknown

Software

The current version runs on any symbolrate, recommended are 5MSymbols/s or more.
You need to set the DAC clock to 7*desired MSymbols per second,because low pixel clocks might not work and high pixel clocks can be scaled down by using multiple pixel per symbol (here 7).

Every symbol is a complex number, that contains an I (real) and Q (imag) value.
By default the signal is lowpass filtered using convolutional sinc function. To edit the impulse response edit the file fragment.glsl. The I/Q values are feed using a unix fifo / a file. 8 byte form two 32bit floats, the first float maps to I, the second to Q. The values have to be between 0.0f and 1.0f. The real part (I) is assigned to the RED and the imag part (Q) to the GREEN channel, while the BLUE channel is set to 0.5f. 8bit and 10bit VGA Graphic cards are supported.

vgatobaseband uses the first VGA output found. If none is found it terminates. The VGA output has to be configured first using xrandr, arandr, ....
The coordinates are saved, a new modeline is added and set. On termination the modeline is removed and the initial mode is restored.
No need to do any modesetting by hand !!!
All baseband signals contains positive as well as negative numbers. If your signal range covers -1.0f to 1.0f you have to convert them.

newval = (float2)(oldval.i/2 + 0.5f, oldval.q/2 + 0.5f)

The newval is now in range 0.0f to 1.0f and can be used with this programm.
Due to most GPU only having 8bit DACs the floats are converted into a range from -127 to 127 (using the blue channel as differential reference).

Arguments

./vgatoiqbaseband [inputfile] [ARGS]

if no inputfile is specified, -t is used

ARGS could be:

-nofilter        disable fragmentshader
-v               be verbose
-verbose         be verbose
-cutofright <n>  cutof n rightmost pixels (hsync)
                 default:  1
-cutofbottom <m> cutof m bottommost pixels (vsync)
                 default:  2
-display DISPLAY Specify the X server to connect to. If not specified, the value of the DISPLAY environment variable is used. 
-direct          force direct rendering
-t               generates test patterns
-pclk <x>        use a pixelclock of x Mhz
                 default: 64
-d <x>           convolutional depth
                 default: 17
-f <c>           same as --freq
--freq <c>       set convolutional filter cut-of frequency to c Mhz (c is float)
                 default: 3.81 Mhz
-h               same as --help
--help           print help message

Examples

Set the VGA position first:

xrandr --output VGA-0 --right-of LVDS --auto

adjust VGA-0 and LVDS according to your system.

./vgatoiqbaseband /tmp/myfifo -v -msps 56 -cutofright 0 -cutofbottom 0

This will read from the file /tmp/myfifo, pixelclk is 56Mhz, it will output 56/7 = 8 MSPS, set HSYNC to 0 and VSYNC to 0.
0 VSYNC might only work on Intel GPU.

./vgatoiqbaseband -v -t

This will generate test paterns, pixelclk is 64Mhz, it will output 64/7 = 9.142 MSPS, set HSYNC to 1 (7 pixels) and VSYNC to 3 lines.
Might work on all GPUs. This also removes the right-most data value in each line and cuts off the 3 bottom-most lines from the input data to prevent phase shifts.

Libraries

  • libglut
  • libglu
  • libgl
  • libxrandr

You may also use

  • libX11

Debian packages:

  • freeglut3-dev
  • libglu1-mesa-dev
  • libgl1-mesa-dev
  • libxrandr-dev

OFDM Bandwidth

OFDM useful carriers to total carriers ratio: 1705 / 2048 = 0.832

DAC clock [Mhz] Msymbols/s carrier ratio 3dB Bandwidth Mhz
64 9.14285 0.832 7.61
56 8.000 0.832 6.66
48 6.8571 0.832 5.71

Phase shift

As phase-shifting OFDM symbols results in malformed spektrum, the horizontal sync is padded with blanking pixel to make sure the sync has the same size as all other symbols.

Interpolation Filter

The convolutional filter is critical in this application. Using OFDM 8 Mhz bandwidth the I and Q channel maximum frequency is 3.81 Mhz. The DAC sampling rate is 9.142 MSPS which is enough to reconstruct all frequencies, but due to the low sampling rate the baseband signal contains aliasing artefakts. To remove those artefakts a si-filter is neccessary, including an interpolation DAC, which is running at much higher sampling rate.
This is done by the fragment shader, implemented in shader.cpp.

The GPU has to fetch,multiply, add and store at least <sampling rate> * <conv_depth> pixels

DAC clock [MSps] GPU Driver max Kernel_size
64 RV710 Mesa 9.0.2 7
64 RV710 fglrx 8.97.2 17
recommended 27 or more

A hardware low-pass filter is required to remove remaining aliasing effects.

Screen size

On fglrx the default screen size is 1600x1600. To increase this limit generate an xorg.conf using 'aticonfig --initial' and add this line:

[...]
Section "Screen"
       Identifier "aticonfig-Screen[0]-0"
       Device     "aticonfig-Device[0]-0"
       Monitor    "aticonfig-Monitor[0]-0"
       DefaultDepth     24
       SubSection "Display"
               Viewport   0 0
               Virtual   3600 1600
               Depth     24
       EndSubSection
EndSection

The line containing "Virtual" is critical. This should be no problem on Mesa as the default screen size is 8196x8196.

TODO

  • Windows modesetting support

Links