Grant (NQ5T) did some careful measurement of the performance of the noise
reduction in the current V2 (2032?) implementation in the Orion II. Although
noise reduction is not the number 1 capability of the Orion DSP for me, I have
gathered some information and data on the noise reduction in the Orion -
running V1.373b5. With the many changes in the DSP firmware it is important to
remember that operation of things like noise reduction can, and often do,
change from release to release. When I say "the Orion" - I mean the Orion
running 1.373b5.
The Orion uses a Least-Mean-Squared algorithm to drive an adaptive filter
structure in the IF chain. One write up about the NR in the Orion said that the
"Multi Knob" selected one of nine different filters - I think that is basically
correct. It also said that once the filter was "formed," even if you changed
the Multi-Knob, the filter did not change. I think that was correct if the
signal in the bandpass does not substantially change. The Orion noise reduction
early documentation indicated that the "aggressiveness" of the filter
adaptation was changed with the Multi Knob. I relate that to the "time to
complete adaptation of the filter." With the multi knob in the 1 position, the
filter is formed quickly. With the Multi Knob at 9 it takes longer for the
filter to be formed. You can see the impact of the knob setting on the filter
design by using SPECTRAN to look at the audio output signal while turning the
NR on and off with different settings on the Multi Knob and looking at
different kinds of signals in the IF bandpass. With the filter in the 1
position it appears that the Orion NR creates either a lowpass or highpass
filter - based on where the significant signal power is. Lowpass if the signal
energy is lower in frequency and highpass if the energy is higher in the
frequency spectrum. When the Multi Knob is set to 2 or above, it appears that
the filter created changes to a bandpass filter - which takes longer (But still
very fast) and uses more DSP cycles. The higher the Multi-Knob setting the
tighter the filter appears to get (Narrower and steeper skirts) when looking at
a steady single frequency signal. When looking at an SSB signal you can see
increased noise reduction in the color (Depicting the level of energy by
frequency) of the waterfall display. The higher the Multi Knob number, the
longer it takes to create the filter. Therefore on rapidly changing signals -
frequency and amplitude - the harder it is for the LMS algorithm to keep u
p with changing the filter as the error signal changes. In many - most? -
cases the multi-knob set to 1, or low numbers, should allow the noise reduction
filter to adapt to SSB signals better with less distortion being created while
the filter adapts (I really hear no distortion with low numbers on the NR
setting in the Orion). Slower changing signals, like CW, can benefit from
higher settings - which create tighter filters around the signal and reduce the
noise to a greater degree. Using SPECTRAN you can see the filter get tighter
around a CW signal as you increase the Multi Knob to 9.
The Orion NR apparently uses a technique which causes the filter to collapse
when the signal is completely removed. This causes the background noise to be
eliminated completely, such as between CW elements.
The Orion noise reduction is within the AGC loop. Therefore when NR is engaged
the noise signals are taken out of the IF bandpass and they do not affect the
AGC level(S Meter drops). If you use a weak signal in the band pass you can
actually see the level of the weak signal increased when the NR is turned on
due to the AGC increasing the gain of the receiver due to the drop in the noise
power. Thus the phrase - the signal popped up out of the noise - noise goes
down and the gain goes up - note the gain being increased doesn't help the
S+N/N - only the noise going down - more then the signal.
Just a couple of observations about NR in the Orion with V2.056. If you look at
the performance of the noise reduction with the new DSP code you will observe
the following things:
1. It appears that only a lowpass filter is ever formed - regardless of the
setting of the Multi Knob. This is true even if the signal energy is in the
upper end of the frequency spectrum. So if the signal of interest has
significant high frequency components, they are attenuated by the filter.
2. It appears that as the Multi-Knob is increased in number, moving towards 9,
the lowpass filter skirt becomes steeper.
3. The NR function is no longer in the AGC loop. Therefore when the NR is
engaged there is no change in the S-Meter/AGC level and if the signal and noise
are reduced the AGC does not increase the gain of the receiver.
4. Engaging NR in V2.056 creates significant attenuation in the total signal
and noise.
Let me say that I know that manually reducing the bandpass for CW operation can
be more effective in improving the S+N/N then the NR of the V1 or V2 code. And,
adjusting the bandpass for the speech frequency of the person you are talking
to on sideband can also be effective in improving the S+N/N (But pretty
awkward). However, I do find the NR on the Orion with V1.373b5 to be helpful in
some operating situations - both on sideband and CW. I also find the
performance of the NR depends a lot on the noise and the signal - you can
also see these effects by looking at SPECTRAN while tuning different signals on
different bands with different background noise. NR is also dependent on other
bandpass adjustments in the Orion. If you narrow the bandpass of the Orion
manually (Reducing the noise bandwidth) the NR algorithm will "starve for
noise" and the filter that is built will be ineffective due to the LMS error
signal going to zero very fast.
Observing signal power levels using the SPECTRAN waterfall display can be very
helpful in seeing the effect of the NR algorithms. Depending on where the
starting bandwidth is, I am confident that the NR algorithm in the Orion
(V1.373b5) can improve the S+N/N for CW signals.
I also believe, from looking at the SPECTRAN data, that the S+N/N can be
improved for SSB signals by using the Orion NR.
I think Grant's (NQ5T) data tells a story that is similar to my observations
with V2.056.
Finally, TenTec is probably the only place where the real data about the Orion
NR is known. I am aware that the same observation can come from different
causes. The data above is from observations. The conclusions could have errors.
73,
Merle - W0EWM
Disclaimer: I like my Orion running V1.373b5, even though it has some stability
issues. I don't plan to sell it. I plan to continue to work with TenTec to
continue to improve the performance and functionality of the Orion V2 DSP code.
_______________________________________________
TenTec mailing list
TenTec@contesting.com
http://lists.contesting.com/mailman/listinfo/tentec
|