Stefan,
Some thoughts related to self-adaptive (set and forget) VLF noise
cancellation [including lightning and cultural (man-made) noise, separately
or together], supported by experiments and literature:
Adaptive neural (echo state, recurrent...) nonlinear blind noise
cancellation techniques can provide 30dB to 50dB improvement in SNR for
signals of known type, even with initial SNR of 0dB, and even with very
complex mixtures of natural and cultural noise. I don't know if any of these
are mature enough for permanent set-and-forget operation, and they can be on
the computationally-heavy side, but:
- They are effective with mixtures of noise that are intractable for
conventional noise-cancellation tools, and:
- They do not require knowledge of the noise characteristics; in fact, to a
considerable extent, the more complex the noise, the better the results; so
it is unnecessary to program characteristics such as rise time, blanking
time, threshold, etc.
There are many publications describing these techniques, from which one can
see that the methods are efficient and effective for signals in (for
example) the VLF frequency range and with noise characteristics such as are
found in the VLF spectrum.
I have been tinkering with intelligent-algorithm cancellation of ULF and VLF
cultural noise for many years, with some success for particular types of
difficult noise such as switchmode noise (motors, power supplies, monitors,
power lines sourcing the above, etc). In the process I have characterized
ULF and VLF noise in many environments, and these characterizations have led
me to believe that despite some success with non-neural/non-adaptive*
intelligent-algorithm cancellation of specific difficult noise types [such
as switchmode converter noise and radiation from power lines sourcing
current to switchmode converters (i.e. most building-wiring and most utility
power lines)], that:
Adaptive neural blind noise cancellation may be a particularly effective
approach for most practical VLF situations, because the multiplicity of
ULF/VLF noise sources with complex time-domain and frequency-domain
characteristics in or near (for example) homes, residential areas, vehicles,
overhead/underground utility power lines and personal electronics, to name a
few, makes successful aggregation of non-neural/non-adaptive*/non-blind
intelligent-algorithm cancellation for each of the difficult noise types
unlikely. The performance of individual non-neural/non-adaptive*intelligent
algorithms focused on specific types of noise tends to degrade when the type
of noise that they are intended to cancel is immersed in complex noises of
other types. For example, the performance of a non-neural/non-adaptive*
intelligent lightning-noise canceller might tend to degrade in the presence
of substantial switchmode-power noise, and vice-versa. Adaptive neural blind
noise cancellation approaches can overcome this limitation while maintaining
reasonable computational burden.
Many adaptive neural blind noise cancellation approaches work better with
aggregations of difficult noise sources than with simple noise profiles; and
have worked well with types of noise and signals found in VLF; and to one of
your specific points, do not require programming of characteristics such as
rise time, blanking time, threshold, etc., and in fact do not require
knowledge of the noise characteristics at all.
* (non-adaptive in the above meaning: cancellation methods in which the
algorithm does not automatically redefine its basic architecture, logical
connectivity, and allocation of computational resources based on ongoing
results)
73,
Jim AA5BW
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of DK7FC
Sent: Thursday, November 3, 2016 9:23 AM
To: [email protected]
Subject: LF: Smart noise cancelling?!?
Hi all,
Last night i thought a bit about noise cancelling on LF/VLF. Depending on
the band and distance and strength of the QRN, different settings for a
noise blanker are used, or optimal. Different rise times, treshold levels
and so on.
I thought about propagation changes and different shapes of QRN bursts in
the time domain, requiring different blanker settings.
Is it possible to program an 'intelligent' noise blanking system that is
evaluating the input spectrum, looking at the shape/type of a sferic and
automatically sets individual dynamic noise blanker parameters for each
burst?
Or do i miss something here?
Just a thought. I guess i'm not the first one who has this idea :-)
73, Stefan
|
