Noise figure paradox
 

On Tue, 24 Mar 2009 23:00:14 -0700 (PDT), JIMMIE
wrote:

On Mar 24, 9:45*pm, Richard Clark wrote:
On Tue, 24 Mar 2009 23:47:52 GMT, "Harold E. Johnson"

wrote:
Deep space communications proceeds many dB below the noise floor
enabled through technology that has become ubiquitous in cell phones -
Spread Spectrum. *I have developed pulsed measurement applications for
which any single pulse has a poor S+N/N, but through repetition
improves S+N/N response with the square root increase of samples
taken.

73's
Richard Clark, KB7QHC

And others call it autocorrelation?

Which?

73's
Richard Clark, KB7QHC

Radar people for one, also known as pulse-pair radar where data from
multiple returns are compared. The data can be from multiple hits on a
target using the same radar or the data can come from multiple radars.
MDS level improvement below the noise level can be achieved. Its also
used for transmitting data.One other specific use I am familiar with
involves transmition of radar data via radio. So the radar uses it as
well as the mode of transmission of the radar data from the radar to
the user.




Jimmie

My question of Which? was directed to Harold's broad brush painting
two different illustrations. Spread spectrum incorporates cross
correlation through slipping the gold code to find a flash. My design
performed a form of forced auto correlation (much like your radar
example, perhaps) but reduced noise as a function of that noise being
uncorrelated to the pulse.

Perhaps this is all saying the same thing at a very fundamental level.
However, I would guess this all hinges on the reduction of noise
following the square root of the ratio of the sample counts.
Conceptually, the distinction between auto or cross correlation is
really of minor consequence.

73's
Richard Clark, KB7QHC

Noise figure paradox
 

Dear Group:

Three conclusions/observations:
1. Noise temperature is the unambiguous way of specifying/describing the
noise performance of a receiver.

2. The paper by Costas in December 1959 Proc of IRE is also valuable to
this discussion. Be sure to read the follow-up comments.

3. I heard with my own ears Shannon observe that, from an engineering point
of view, if one did not have an occasional transmission error one was using
a wasteful amount of power. Shannon was a Michigan boy. 60 dB SNR??? Not
in fly-over land.

73, Mac N8TT

--
J. McLaughlin; Michigan, USA
Home:
"Richard Clark" wrote in message
...
On Fri, 20 Mar 2009 19:46:53 -0700, "Joel Koltner"
wrote:

Say I have an antenna that I know happens to provide an SNR
of 60dB...

Returning to one of the few quantifiables, it would be instructive to
judge why it is so astonishing as a point to begin a dive into the
discussion of noise figure. In other posts related to deep space
probe's abilities to recover data from beneath the noise floor, much
less cell phones to operate in a sea of congestion, I encountered the
economic objection that such methods cost too much - expense of
bandwidth.

Well, not having seen anything more than yet another qualification -
how much is "too much?" It is time to draw back and ask how much is
enough? What would NOT be too expensive? Replacing qualitative
objections with quantitative objections sometimes evokes a horse laugh
when the magnitude of the qualitative issue ceases to exhibit much
quality.

I won't open this round of enquiry with exotic Spread Spectrum which
portends the objection of phase issues with clocks (even knowing that
such modulation techniques automatically incorporate slipping to
adjust for just such problems). Instead I will slip back some 60
years to the seminal paper published by Claude Shannon who figured
this all out (with H.W. Bode) and quote some metrics for various
coding (modulation) schemes. Search for "Communication in the
Presence of Noise." When you google, search in its image data space
for the cogent chart that I will be drawing on, below. Obtaining the
paper may take more effort (or simply email me for a copy).

Starting with BPSK and a S+N/N of roughly 10.5 dB, the bit error rate
is one bad bit in one million bits. This is probably the most
plug-ordinary form of data communication coming down the pike; so one
has to ask:
"is this good enough?"
If not, then "SNR of 60dB" is going to have to demand some really
astonishing expectations to push system designers to ante up the
additional 49.5 dB.

Well, let's say those astonishing expectations are as wild as
demanding proof that you won't contribute to global warming if you
chip an ice cube off of a glacier - such are the issues of scale when
you chug the numbers for BPSK.

OK, so as to not melt down the planet, we step up the complexity of
modulation to better than the last solution for "good enough." Let's
take the Voyager probes of the deep planets where at a S+N/N of 2.53
dB (in what is called 8 dB coding gain) the same error rate of 1 bit
in 1 million is achieved. One has to ask:
"is this good enough?"
If not, then "SNR of 60dB" is going to have to demand some really
astronomical expectations.

OK, perhaps this is a problem demanding really deep pockets that
exceed the several $Trillion being spent on the past 8 years of
Reaganomic neglect. (Why else pound the desk for that extra 57 dB?)
Let's go the full distance to the Shannon limit. It will give us that
same 1 bit error for every 1,000,000 at -1.5 dB S+N/N. If this isn't
below the noise floor, then the problem demanding 60 dB will never
find the solution to positively answer:
"is this good enough?"

73's
Richard Clark, KB7QHC

Noise figure paradox
 

Hi Richard,

"Richard Clark" wrote in message
...
That was a curious objection to a solution answering a problem as it
was specifically stated. Are there angles to showing noise being
overcome by several means when you offered none?

My means were "reduce the noise figure of the amplifiers in your front-end"
and "reduce the phase noise of your oscillators/PLLs/etc." "Averaging the
input" is a clear winner here too.

What "noise" were you speaking about when through the course of this
thread it has most often been confined to kTB than, say, cross-talk,
splatter, spurs, whistlers, howlers, jamming, and a host of others?

For the sake of this thread, it's been just thermal and oscillator nose since
these are -- AIUI -- what limit traditional analog (AM/FM/PM) communication
systems. Most of the rest of what you've listed are certainly real-world
problems, but they're (hopefully) somewhat transient in nature and -- as
AIUI -- often all lumped into a single "fade margin" when designing the an
end-to-end system. E.g., the transmission medium is often modeled with
something no more complex than, say, the Friis equation and Rayleigh fading.
I do realize that in the real world things like spurs or splatter can end up
being very expensive (frequency changes, high-order/high-power filters, etc.)
if you're co-locating your radio with many others on a hilltop -- I've been
told that if you take a run-of-the-mill radio to a place like Mt. Diablo in
California, many of them just fall over from front end overload and cease to
function at all.

What constitutes "successfully?" Is this a personal sense of well
being, or is it supported by a metric?

Usually something like a 12dB SINAD standard is used for analog modulation
schemes or a 1e-3 bit-error rate for digital modulation techniques (before any
error correction coding is applied).

Spread Spectrum is so ubiquitous that waiting on anticipated exotic
failures of phase noise, on the face of an overwhelming absence of
problems, is wasted time indeed.

It's not ubiquitous on amateur radio, though.

But yeah, commercially it certainly is, and my understanding is that phase
noise in oscillators in a Big Deal for cell sites, requiring much more
strigent standards than what a 2m/440 HT's oscillator is likely to provide.
The network timing of cell sites is sync'd to atomic clocks via
GPS-disciplined oscillators system as well.

As to sampling error via the net. Time was when 16x over-sampling for
RS-232 was the norm.

I've meet many RS-232 routines that don't do any over-sampling at all -- I've
even written a few. :-) For most applications the SNR of an RS-232 signal is
typically well in excess of 20dB if you don't exceed the original specs for
cable length of bit rate. (Granted, as least historically before RS-232
starting falling out of use, it was probably one of the most "abused"
electrical interconnect standards in the world, and 16x oversampling certainly
would let you go further than simple-minded receivers with no oversampling.)

---Joel

Noise figure paradox
 

Hi Richard,

"Richard Clark" wrote in message
...
In other posts related to deep space
probe's abilities to recover data from beneath the noise floor, much
less cell phones to operate in a sea of congestion, I encountered the
economic objection that such methods cost too much - expense of
bandwidth.

I don't think anyone stated they cost "too much," just that there is a cost in
increased bandwidth, and bandwidth isn't free.

In general the spread spectrum processing gain is proportional to the
bandwidth increase over what the original data stream would require without
any spreading.

Well, not having seen anything more than yet another qualification -
how much is "too much?"

Definitely depends on "the market." You can bet the cell phone developers
have sophisticated models of possible radios and the channel and associate
with each piece a cost (e.g., bandwidth = $xx/Hz, improving close-in phase
noise of master oscillator = $xx/dBc, etc.), and then run a lot of simulations
to try to make the average cost of each bit as low as possible. Of course,
there are many variables that are impossible to ascertain precisely such as
how quickly uptake of new cell services (e.g., 3G data) will be in a given
area (as this drives how many towers you put there initially and how quickly
you roll out more), how fast fab yields will improve that lower your costs and
improve RF performance, etc.

Starting with BPSK and a S+N/N of roughly 10.5 dB, the bit error rate
is one bad bit in one million bits. This is probably the most
plug-ordinary form of data communication coming down the pike; so one
has to ask:
"is this good enough?"
If not, then "SNR of 60dB" is going to have to demand some really
astonishing expectations to push system designers to ante up the
additional 49.5 dB.

Why Richard, I'm starting to think you don't spend thousands of dollars per
meter on your speaker cables. :-) Hey, see this:
http://www.noiseaddicts.com/2008/11/...st-audiophile/ -
- $7,000/m speaker cables! Includes, "Spread Spectrum Technology!" :-)

That being said, back in the analog broadcast TV days (oh, wait, not all of
them are gone yet, but they will be soon), I believe that "studio quality"
NTSC is considered to be 50dB SNR (for the video), whereas people would start
to notice the noise if the received signal's SNR had dropped below 30ish dB,
and 10dB produces an effectively unwatchable pictures. This reinforces your
point that "good enough" is highly subjective depending on how the
"information" transmitted is actually used.

You make a good point that the Shannon limit gives a good quantitative measure
of how you go about trading off bandwidth for SNR (effectively power if your
noise if fixed by, e.g., atmospheric noise coming into an antenna). Shannong
doesn't give any hint as to how to achieve the limits specified, although I've
read that with fancy digital modulation techniques and "turbo"
error-correcting codes, one can come very close to the limit.

---Joel

Noise figure paradox
 

"J. Mc Laughlin" wrote in message
.. .
2. The paper by Costas in December 1959 Proc of IRE is also valuable to
this discussion. Be sure to read the follow-up comments.

Is that available publicly anywhere?

3. I heard with my own ears Shannon observe that, from an engineering point
of view, if one did not have an occasional transmission error one was using
a wasteful amount of power. Shannon was a Michigan boy. 60 dB SNR??? Not
in fly-over land.

I think the counterpoint is that, particularly in mobile environments, you
often needed huge fade margins, e.g., 20-40dB wasn't uncommon for pager
systems. Hence in systems designed to have, say, an "average" of 30dB SNR
(same audio quality as the telephone system, assuming 3kHz bandwidth as well),
it wouldn't be surprising to occasionally find you're actually getting 60dB
SNR in the most ideal scenario.

Although perhaps designing for an average of 30dB SNR is a little high for a
paging system... anyone know? (I'm thinking 20dB might be a bit more
realistic.)

---Joel

Noise figure paradox
 

"Joel Koltner" wrote in message
...
Is that available publicly anywhere?

What I really meant here was, "Is that available *to download from the
Internet* publicly anywhere?"

Noise figure paradox
 

That's great information, Jim, thanks!

Noise figure paradox
 

Hi Richard,

"Richard Clark" wrote in message
...
I don't think anyone stated they cost "too much," just that there is a cost
in
increased bandwidth, and bandwidth isn't free.
Um, this last statement seems to be hedging by saying the same thing
in reverse order.

No, they really are different. What costs too much for me might very not cost
too much for the military or NASA, for instance.

It would be more compelling if you simply stated the cost for ANY
market.

The original example was meant to be more of a "textbook" problem, hence the
lack of elaboration on the specifics of the "market" involved.

I would suspect that "studio quality" observes other characteristics
of the signal.

Agreed, I would too.

A multipath reception could easily absorb a
considerable amount of interfering same-signal to abyssmal results. It
would take a very sophisticated "noise" meter to perform the correct
S+N/N.

Yep, very true -- I think this is why you see people legtimately complaining
about the quality of their cable TV even though the cable installation tech
whips out his SINAD meter and verifies it meets spec; the quality of a
transmission can't always be boiled down to just one number.

The "Turbo" codes are achievable in silicon with moderate effort. A
work going back a dozen years or more can be found at:
http://sss-mag.com/G3RUH/index2.html

Great link, thanks!

---Joel

Noise figure paradox
 

On Wed, 25 Mar 2009 11:11:48 -0700, "Joel Koltner"
wrote:

Hi Richard,

"Richard Clark" wrote in message
.. .
In other posts related to deep space
probe's abilities to recover data from beneath the noise floor, much
less cell phones to operate in a sea of congestion, I encountered the
economic objection that such methods cost too much - expense of
bandwidth.

I don't think anyone stated they cost "too much," just that there is a cost in
increased bandwidth, and bandwidth isn't free.

Um, this last statement seems to be hedging by saying the same thing
in reverse order.

Well, not having seen anything more than yet another qualification -
how much is "too much?"

Definitely depends on "the market."

It would be more compelling if you simply stated the cost for ANY
market. Qualified statements are suitable for Madison Avenue to sell
cheese, but it doesn't make for an informed cost-based decision.

That being said, back in the analog broadcast TV days (oh, wait, not all of
them are gone yet, but they will be soon), I believe that "studio quality"
NTSC is considered to be 50dB SNR (for the video), whereas people would start
to notice the noise if the received signal's SNR had dropped below 30ish dB,
and 10dB produces an effectively unwatchable pictures. This reinforces your
point that "good enough" is highly subjective depending on how the
"information" transmitted is actually used.

I would suspect that "studio quality" observes other characteristics
of the signal. A multipath reception could easily absorb a
considerable amount of interfering same-signal to abyssmal results. It
would take a very sophisticated "noise" meter to perform the correct
S+N/N.

You make a good point that the Shannon limit gives a good quantitative measure
of how you go about trading off bandwidth for SNR (effectively power if your
noise if fixed by, e.g., atmospheric noise coming into an antenna). Shannong
doesn't give any hint as to how to achieve the limits specified, although I've
read that with fancy digital modulation techniques and "turbo"
error-correcting codes, one can come very close to the limit.

The "Turbo" codes are achievable in silicon with moderate effort. A
work going back a dozen years or more can be found at:
http://sss-mag.com/G3RUH/index2.html
(consult the adjoining pages for fuller discussion)

73's
Richard Clark, KB7QHC

Noise figure paradox
 

"Richard Clark" wrote in message
...
Which is no more complex than setting 4 register bits - I wouldn't
call that a "routine," however.
-- I've
even written a few. :-)
Why more than one? Were the rest undersampling routines?

These were software RS-232 receivers, so you make use of whatever timers, edge
interrupts, etc. that you have sitting around to first the start bit, load up
a timer to then trigger in (what should be) the middle of the bit time for the
sample, etc. I've written pretty much the same routines a small handful of
times on different CPUs and in different languages.

The first ones I wrote were on ~1MIP CPUs in assembly and were limited to
about 2400bps full-duplex if you were also trying to run a reasonably
responsive terminal emulator (e.g., wanted to still have 50% of the CPU
available for the emulator), whereas more recently I've written them on ~20MIP
CPUs in C and can easily do 9600bps full-duplex with only a small impact on
CPU usage.

---Joel