Random Source: Difference between revisions

From GNU Radio
Jump to navigation Jump to search
(Rectifying misleading random number quality)
No edit summary
 
(4 intermediate revisions by 2 users not shown)
Line 1: Line 1:
[[Category:Block Docs]]
[[Category:Block Docs]]
Generates a number of samples of random numbers of [min, max) meaning the max value won't be included. Repeat samples if specified.  Great for creating bytes of information for a modulator.
Generates a number of samples of random numbers of [min, max) meaning the max value won't be included. Repeat samples if specified.  Useful for creating bytes of information for testing a modulator.


Ex: With min=0 and max=2, the sequence 01110101... will be generated.
Ex: With <code>min=0</code> and <code>max=2</code>, the sequence 01110101... of length <code>num_samps</code> will be generated.


Supports an output of type int, short, and byte.
Supports an output of type int, short, and byte.
Line 8: Line 8:
This block differs from [[Random Uniform Source]]:  
This block differs from [[Random Uniform Source]]:  


This block uses Numpy to generate a fixed random vector of values. The output of this block repeats every <code>num_samples</code>, and hence has high autocorrelation with that period.<br>
This block uses Numpy to generate a fixed random vector of values. The output of this block repeats every <code>num_samps</code>, and hence has high autocorrelation with that period.<br>
The algorithms used to generate the random numbers differ, as well: Numpy, and hence Random Source, use https://en.wikipedia.org/wiki/Mersenne_Twister MT19937] (but the slightly superior randomness properties don't matter within the maximum size of a vector of values on a PC), whereas the [[Random Uniform Source]] uses https://xoroshiro.di.unimi.it/xoroshiro128plus.c XOROSHIRO128+], which has a period of 2<sup>128</sup>-1 and is hence to be preferred if autocorrelation over a window of length <code>num_samps</code> is relevant.
The algorithms used to generate the random numbers differ, as well: Numpy, and hence Random Source, use [https://en.wikipedia.org/wiki/Mersenne_Twister MT19937]¹, whereas the [[Random Uniform Source]] uses [https://xoroshiro.di.unimi.it/xoroshiro128plus.c XOROSHIRO128+], which has a period of 2<sup>128</sup>-1 and is hence to be preferred if autocorrelation over a window of length <code>num_samps</code> is relevant.
 
This block does not support C++ output, so it cannot be used when the output language of a flowgraph in GRC is C++.


== Parameters ==
== Parameters ==
(''R''): <span class="plainlinks">[https://wiki.gnuradio.org/index.php/GNURadioCompanion#Variable_Controls ''Run-time adjustable'']</span>
(''R''): <span class="plainlinks">[https://wiki.gnuradio.org/index.php/GNURadioCompanion#Variable_Controls ''Run-time adjustable'']</span>


; Param 1 (''R'')
; Output type
: Description of parameter, provide any tips or recommended values.  Note that the name of the parameter above should match the param's label that shows up in grc (e.g. Sample Rate).
: Available options are int, short or byte
 
; Minimum
: The lower limit of the range of generated values (included in the output)
 
; Maximum
: The upper limit of the range of generated values (not included in the output)


; Param 2
; Num Samples
: blah blah blah
: Total number of samples that are generated in the output
 
; Repeat
: Yes/No


== Example Flowgraph ==
== Example Flowgraph ==
Line 36: Line 47:


; Block definition
; Block definition
: [https://github.com/gnuradio/gnuradio TODO]
: [https://github.com/gnuradio/gnuradio/blob/main/gr-analog/grc/analog_random_source_x.block.yml analog_random_source_x.block.yml]
 
 
== Computational Performance ==
 
[[File:Benchmark prng.grc|alt=Benchmarking]] done on a Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz, 8GB RAM VOLK 2.5.0, GNU Radio 3.10.0.0-rc1, Linux 5.15.11
 
[[File:Benchmark prng.png]]
 
{| class="wikitable"
|-
! Block
! Number of Samples / Period after which samples repeat
! Throughput (in 10<sup>6</sup> samples per second)
! Note
|-
| Random Source
| 1000
| 838
| default setting
|-
| Random Source
| 2<sup>20</sup>=1,048,576
| 824
|-
| Random Source
| 2<sup>24</sup>
| 812
|-
| Random Source
| 2<sup>28</sup>
| 813
| Initialization of the random vector needs nearly 30s, eats 4GB of RAM
|-
|
|-
| Random Uniform Source
| 2<sup>128</sup>-1
| 282
|
|}
 
 
----
 
¹ The slightly superior randomness properties of MT19937 over XOROSHIRO128+ don't matter within the maximum size of a vector of values on a PC, and are generally unnoticable in an SDR contex

Latest revision as of 05:52, 20 January 2022

Generates a number of samples of random numbers of [min, max) meaning the max value won't be included. Repeat samples if specified. Useful for creating bytes of information for testing a modulator.

Ex: With min=0 and max=2, the sequence 01110101... of length num_samps will be generated.

Supports an output of type int, short, and byte.

This block differs from Random Uniform Source:

This block uses Numpy to generate a fixed random vector of values. The output of this block repeats every num_samps, and hence has high autocorrelation with that period.
The algorithms used to generate the random numbers differ, as well: Numpy, and hence Random Source, use MT19937¹, whereas the Random Uniform Source uses XOROSHIRO128+, which has a period of 2128-1 and is hence to be preferred if autocorrelation over a window of length num_samps is relevant.

This block does not support C++ output, so it cannot be used when the output language of a flowgraph in GRC is C++.

Parameters

(R): Run-time adjustable

Output type
Available options are int, short or byte
Minimum
The lower limit of the range of generated values (included in the output)
Maximum
The upper limit of the range of generated values (not included in the output)
Num Samples
Total number of samples that are generated in the output
Repeat
Yes/No

Example Flowgraph

Random-source-ex.png

Source Files

C++ files
TODO
Header files
TODO
Public header files
TODO
Block definition
analog_random_source_x.block.yml


Computational Performance

File:Benchmark prng.grc done on a Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz, 8GB RAM VOLK 2.5.0, GNU Radio 3.10.0.0-rc1, Linux 5.15.11

Benchmark prng.png

Block Number of Samples / Period after which samples repeat Throughput (in 106 samples per second) Note
Random Source 1000 838 default setting
Random Source 220=1,048,576 824
Random Source 224 812
Random Source 228 813 Initialization of the random vector needs nearly 30s, eats 4GB of RAM
Random Uniform Source 2128-1 282


¹ The slightly superior randomness properties of MT19937 over XOROSHIRO128+ don't matter within the maximum size of a vector of values on a PC, and are generally unnoticable in an SDR contex

Retrieved from "https://wiki.gnuradio.org/index.php?title=Random_Source&oldid=10331"