Re: how to generate a certain band width white noise

[Doug McGarrett]

On 3/22/21 1:08 AM, Kyeong Su Shin wrote:
> Hello James:
>
> The bandwidth of the "Noise Source" block is 2pi. It's a digital 
> signal, so the bandwidth is not in MHz, and is in range of [0, 2pi] or 
> [0, 1] (if you normalized it).
>
> You can calculate an equivalent analog bandwidth if you know the 
> sampling rate of hardware sink block that is being used (ex: UHD sink, 
> Osmocom Sink, etc). If the sampling rate is set to 4 MS/s and if a 
> Noise Source block is connected to the hardware sink directly, then 
> the bandwidth of the outputted noise should be (ideally) 4 MHz. If you 
> use a resampler to interpolate the signal, it will go down. 
> Alternatively, you can also use a low-pass filter to reduce the 
> bandwidth (forgot to mention this in the previous e-mail, but it is 
> probably faster and acceptible in this case).
>
> Regards,
> Kyeong Su Shin
> ------------------------------------------------------------------------
>
> <http://aka.ms/weboutlook>
>
> ------------------------------------------------------------------------
For a truly white noise source, I believe that you need a truly white 
noise source! That may seem redundant, but what I'm suggesting
is a device made to produce white noise--in most cases a particular kind 
of diode, biased by a DC current and followed by a known
amount of attenuation in order to produce a known white noise level, 
specified as "excess noise" over a given bandwidth--usually
from a frequency in the HF band to somewhere in the microwave range, 
frequently a frequency above 1 GHz. This device is expensive,
and is subject to calibration to NIST standards. There have been some 
articles in the Amateur Radio magazines over the years
on building your own noise source, but the calibration of such is not 
guaranteed.
If you find a noise head--most commonly a male type N output with a BNC 
input--it will require +28 volts DC on  the BNC connector.
The most commonly available type produces ~15.4 dB excess noise, but 
other values are also available. When measuring
a receiver with a very low noise figure--one in the range of less than 
1dB to maybe 3 or 4 dB, greater accuracy may be  obtained
with a noise source that produces a level of about 5 dB excess noise. 
(Where "about" is a calibrated value.)
To get this noise in a particular bandwidth, you would need a 50 Ohm 
bandpass filter at the type N output port. The excess noise
level would then be determined by the insertion loss of the filter in 
the desired bandwidth.
*I would think that there probably is no reason to limit the bandwidth 
of the noise source at its origin; the receiver under test will**
**have its own filtering internally. *
To determine the noise figure of the receiver or amplifier under test, 
the noise out of the device is measured with the noise source
not powered on, and then with the noise source powered on. This  is 
optimally measured with an RF low-level power meter or
some other device which is accurately calibrated and responds to the 
noise power.
There are several methods of determining the noise figure of the device 
under test. A good write-up is found here:
https://www.maximintegrated.com/en/design/technical-documents/tutorials/2/2875.html
Where reference to a spectrum analyzer is found, one of the SDR units 
can be used in its spectrum situation. Note that the
noise figure of the second device--in this case, a spectrum 
analyzer--must be factored into the equations for calculating the
noise figure of a preamplifier. For an entire receiver, the Y-factor 
method is probably the simplest measurement.