[BC] IBOC "secrets" and my opinions.

[Robert Orban]

At 05:08 AM 3/28/2007, Richard Johnson  wrote:

>On Tue, 27 Mar 2007, Robert Orban wrote:
>
> > At 10:39 AM 3/27/2007, Richard Johnson wrote:
> >
> > Just a few comments:
> >
> >> For the technical issues, the problem with digital and AM within the
> >> same spectrum space is that the digital bins (samples) need to fit in
> >> between the AM bins. In the case of digital, the difference between a
> >> wanted bin and an interfering (unwanted) bin needs to only be that the
> >> wanted bin be greater. Certainly a ratio of 2:1 (6 dB) would result in a
> >> completely clean digital signal. However, for the AM of conventional
> >> design (peak detector), the ratio ends up being the same as the
> >> S/N, i.e., 100:1 for 40 dB, 400:1 for 52 dB, etc., when you
> >> use envelope detection. Digital detection with filtering works by
> >> completely suppressing __anything__ that is not in the passband.
> >> This means that anything removed from the carrier by more than the
> >> audio passband of 10 kHz is gone. It is removed __before__ any of the
> >> nonlinear effects of analog circuitry can occur. The nonlinearity is
> >> the nonlinearity of the A/D converter at the head-end.
> >
> > No digital filter can completely suppress out of band power. Theoretical
> > "brick wall" filters require infinite time delay, and this is true
> > regardless of the technology implementing the filter -- it's a "laws of
> > physics" issue. Digital filters have cost/performance tradeoffs like any
> > other technology. They can be very effective, and the cost for making very
> > selective filters keeps decreasing. But "completely suppressing?" Can't be
> > done.
> >
>
>Well yes it can. It was described in my first post on this subject.
>I also described a 16-bit 20 MHz system so the 96 dB was previously
>qualified. "Completely eliminated," in this context means that there
>are no bits exercised in the 16-bit system. It is done by doing a
>DFT or FFT, zeroing out the offending bins, then performing an inverse.
>To obtain sufficient resolution, one needs to use a decent window
>(a slightly modified Bartlett window will do) and after all is done,
>the result needs to be convolved with the same window. This is all
>possible because, as previously described, "realtime" is only 10,000
>events per second.

Using an FFT for fast convolution is just a computationally efficient way 
to implement an FIR filter at the expense of adding throughput delay to the 
system. The resulting filter will still have finite transmission in its 
stopband. If you are saying that the out of band power can be filtered so 
that it is below the numerical noise, then I agree. But if dither was 
properly used to linearize the A/D conversion, then there will still be out 
of band power present below the noise floor, albeit at a level that is only 
of academic interest. (In a correctly dithered system, the LSB is always 
being exercised by the dither noise.)

(This issue is related to the digital audio myth that there is no 
information below the LSB. This is only true if the system has not been 
correctly dithered. In a correctly dithered 16-bit system, for example, it 
is quite possible to detect a sinewave at -110 dBfs by using sufficiently 
narrow bins in a spectrum analyzer. Our ears have narrowband filters 
built-in, and it is likely that this -110 dBfs sinewave would be audible.)

> > Moreover, DSP technology for consumer radios (including digital IF
> > filtering and detection) has been around for some time and available for
> > purchase in some currently manufactured radios. Eventually, I expect that
> > all radios will be DSP-based because the tech will get progressively
> > cheaper as time goes on.
> >
>
>If are the same Orban that makes the digital stereo generators and
>Optimod (I've been out of the broadcast industry for 30 years), then
>you know that you can even generate a "perfect" FM stereo signal
>with zero 38 kHz (completely suppressed) using modern technologies.

This is essentially true. although some negligible 38 kHz may arise due to 
numerical noise in upstream IIR filters, which can cause very small 
signal-dependent DC offsets between the left and right channels.

Bob Orban