Aybody have a good reference/tutorial on noise and double ba

[John Larkin]

On Tue, 15 Apr 2014 14:37:53 -0700, Jeff Liebermann <jeffl@cruzio.com>
wrote:

On Mon, 14 Apr 2014 21:35:50 -0700, John Larkin
jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:


tried this? http://www.saturnpcb.com/pcb_toolkit.htm
-Lasse

It's always fun to compare microstrip calculators, especially at high w/h
ratios.

Saturn only allows w/h up to 3, so I tried that. W=30 mils, T=10, Er=4.6.
Saturn 29.42 ohms
Appcad 35.70 ohms
Txline 35.71 ohms

Saturn may be using one of the old classical microstrip formulas. They get bad
at high w/h, and many go negative. There are some really bad javascript ones
online.

Methinks you might be referring to mcalc, which the author admits has
problems and was replaced by wcalc:
http://wcalc.sourceforge.net

Actually, I was referring to the type of formula that was in the
Motorola ECL handbook, and in similar semi appnotes.

Any single formula for microstrip impedance will be valid over a
limited range of geometries. Wide traces made negative impedances in
the Moto formula.


--

John Larkin Highland Technology, Inc

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[RobertMacy]

On Mon, 14 Apr 2014 21:35:50 -0700, John Larkin
<jjlarkin@highnotlandthistechnologypart.com> wrote:

...snip....

It's always fun to compare microstrip calculators, especially at high w/h
ratios.

Saturn only allows w/h up to 3, so I tried that. W=30 mils, T=10, Er=4.6.

Saturn 29.42 ohms

Appcad 35.70 ohms

Txline 35.71 ohms

Saturn may be using one of the old classical microstrip formulas. They
get bad
at high w/h, and many go negative. There are some really bad javascript
ones
online.

interesting

I tried old Dimstrip from the DOS days and got 36.34 ohms, remember the
RF/microwave engineers used to claim 1% accurate?

then went to a crude femm analysis at 1MHz and got 37.94
we're talking a 6% error, but what's correct? I'd go to the PCB ouses

 

[whit3rd]

On Wednesday, April 16, 2014 9:43:34 AM UTC-7, Phil Hobbs wrote:
[on modelling transmission lines in printed wiring]

With ordinary FR4, epsilon can range from about 3.8 to 4.5, so your
trace impedances are going to be uncertain by about

deltaZ/Z ~ 1-sqrt((1+3.8)/(1+4.5)) ~ 7%

anyway. That's another reason to ask the board house.

And, it's another reason to include a test coupon! The board
house SHOULD know, but they're certainly as reliable as the
watchful eye of their customers demands...

 

[Phil Hobbs]

On 04/16/2014 10:05 AM, RobertMacy wrote:

On Mon, 14 Apr 2014 21:35:50 -0700, John Larkin
jjlarkin@highnotlandthistechnologypart.com> wrote:

...snip....

It's always fun to compare microstrip calculators, especially at high w/h
ratios.

Saturn only allows w/h up to 3, so I tried that. W=30 mils, T=10, Er=4.6.

Saturn 29.42 ohms

Appcad 35.70 ohms

Txline 35.71 ohms

Saturn may be using one of the old classical microstrip formulas. They
get bad
at high w/h, and many go negative. There are some really bad
javascript ones
online.


interesting

I tried old Dimstrip from the DOS days and got 36.34 ohms, remember the
RF/microwave engineers used to claim 1% accurate?

then went to a crude femm analysis at 1MHz and got 37.94
we're talking a 6% error, but what's correct? I'd go to the PCB ouses

With ordinary FR4, epsilon can range from about 3.8 to 4.5, so your
trace impedances are going to be uncertain by about

deltaZ/Z ~ 1-sqrt((1+3.8)/(1+4.5)) ~ 7%

anyway. That's another reason to ask the board house.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

 

[John Larkin]

On Wed, 16 Apr 2014 12:43:34 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 04/16/2014 10:05 AM, RobertMacy wrote:
On Mon, 14 Apr 2014 21:35:50 -0700, John Larkin
jjlarkin@highnotlandthistechnologypart.com> wrote:

...snip....

It's always fun to compare microstrip calculators, especially at high w/h
ratios.

Saturn only allows w/h up to 3, so I tried that. W=30 mils, T=10, Er=4.6.

Saturn 29.42 ohms

Appcad 35.70 ohms

Txline 35.71 ohms

Saturn may be using one of the old classical microstrip formulas. They
get bad
at high w/h, and many go negative. There are some really bad
javascript ones
online.


interesting

I tried old Dimstrip from the DOS days and got 36.34 ohms, remember the
RF/microwave engineers used to claim 1% accurate?

then went to a crude femm analysis at 1MHz and got 37.94
we're talking a 6% error, but what's correct? I'd go to the PCB ouses


With ordinary FR4, epsilon can range from about 3.8 to 4.5, so your
trace impedances are going to be uncertain by about

deltaZ/Z ~ 1-sqrt((1+3.8)/(1+4.5)) ~ 7%

anyway. That's another reason to ask the board house.

Cheers

Phil Hobbs

Board houses also want some tolerance on the dielectric thickness.

You can pay them to target the actual trace impedances, but that can
get complex and expensive.

I usually hit about +-10% with reasonable tolerances and just asking
for FR4. That's good enough for most applications. It's best to try to
design so that it doesn't matter much. That's easy for digital boards.




--

John Larkin Highland Technology, Inc

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[John Larkin]

On Wed, 16 Apr 2014 11:37:34 -0700 (PDT), whit3rd <whit3rd@gmail.com>
wrote:

On Wednesday, April 16, 2014 9:43:34 AM UTC-7, Phil Hobbs wrote:
[on modelling transmission lines in printed wiring]

With ordinary FR4, epsilon can range from about 3.8 to 4.5, so your
trace impedances are going to be uncertain by about

deltaZ/Z ~ 1-sqrt((1+3.8)/(1+4.5)) ~ 7%

anyway. That's another reason to ask the board house.

And, it's another reason to include a test coupon! The board
house SHOULD know, but they're certainly as reliable as the
watchful eye of their customers demands...

I often include a couple of SMA footprints on a board, and connect
them with a 50 ohm trace that hops through all the layers. I can
install the connectors and TDR that and see how well I did on trace
impedances.

https://dl.dropboxusercontent.com/u/53724080/TDR/Chimera_TDR/Chimera_Test_Trace.jpg

https://dl.dropboxusercontent.com/u/53724080/TDR/Chimera_TDR/Chimera_SD24.JPG

https://dl.dropboxusercontent.com/u/53724080/TDR/Chimera_TDR/Chimera_TDR0.JPG

You can see the connector transition and the vias, too. The 30 ps
scope rise time gets trashed by the trace losses.

https://dl.dropboxusercontent.com/u/53724080/TDR/Chimera_TDR/TDT_risetime.JPG



--

John Larkin Highland Technology, Inc

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Simon S Aysdie]

On Monday, April 14, 2014 7:28:53 AM UTC-7, Robert Macy wrote:

On Sun, 13 Apr 2014 23:45:10 -0700, Tim Wescott <tim@seemywebsite.please
wrote:
...snip...

Wes Hayward, "Introduction to Radio Frequency Design", Prentice Hall,
1982.

There's lots of different kinds of double balanced mixers. I don't know
much about the others, but a diode ring mixer basically does nothing to
the noise, but cuts the signal amplitude by a factor of 2. So it adds
6dB to the noise figure of your signal processing chain.

Thanks for the reference.

The Hayward book is out of print, btw. Good book.

Diode rings generate EVERY ODD harmonic of the carrier, which because they
do NOT get filtered out, will get in the way. I know, I know. People give
a solution and I respsond with, "Oh, thre's one more thing" I hate that
too. Then again, *if* diode rings can be used, they have to be
cheaper/simpler, maybe there's still some way to do that. Thanks for the
'food for thought'

I still wonder why you can't filter them out. This is usually "easy," so this naturally leads to "why not?"

Has some treatment of noise:
Fundamentals of Mixer Design, Agilent, Steve Long
http://cp.literature.agilent.com/litweb/pdf/5989-9102EN.pdf

other mixer stuff
-----------------
Well known papers, but didn't find online:
Sources of intermodulation in diode-ring mixers, H. P. WALKER
Predicting Intermodulation Suppression in Double-Balanced Mixers, Bert Henderson, WJ Tech Note
The Relationship Between Cross-Modulation and Intermodulation Distortions in the Double-balanced Modulator, JOHN G. GARDINER

(The Walker paper is cited a great deal.)

Easy hits:
http://www-atom.fysik.lth.se/QI/laser_documentation/Selected_articles/wj%20Mixers_part1.pdf
http://www-atom.fysik.lth.se/QI/laser_documentation/Selected_articles/wj%20Mixers_part_2.pdf
http://rfcafe-com.secure38.ezhostingserver.com/references/articles/wj-tech-notes/ImageRej_n_SSB_mixers.pdf
http://www.rfcafe.com/references/articles/wj-tech-notes/Mixers_in_systems_part1.pdf
http://rfcafe-com.secure38.ezhostingserver.com/references/articles/wj-tech-notes/Mixers_in_systems_part2.pdf

http://michaelgellis.tripod.com/mixersin.html
http://www.qsl.net/va3iul/RF%20Mixers/RF_Mixers.pdf
http://www.minicircuits.com/app/AN00-009.pdf

 

[Simon S Aysdie]

On Monday, April 14, 2014 6:16:14 AM UTC-7, Gerhard Hoffmann wrote:

> < http://arxiv.org/abs/physics/0608211 >

Nice. Thanks.

 

[RobertMacy]

On Thu, 17 Apr 2014 11:44:34 -0700, Simon S Aysdie <gwhite@ti.com> wrote:

...snip...
I still wonder why you can't filter them out. This is usually "easy,"
so this naturally leads to "why not?"

The carrier frequency is NOT fixed, and could go low enough that over 50
harmonics are lower than that first fundamental. Not easy to filter that
out. Tracking filters? can I get more thsn 60dB rejection to the third?
I'd like 80 dB. Any upper harmonic slipping in causes problems. Perhaps,
tracking filter will be equivalent. I like the idea of using hard driven
diode ring mixers for their simplicity. just have to add that elusive
tracking filter to rid of anything much above the fundamental. If this
were audio, I see a way to do it, but then again if audio, wouldn't have
to, multipliers exist.

...snip ... EXCELLENT list of references!

Thank you! Now I know what it feels like to take a drink from a firehose!

What do you work in? TI email address? does TI make a high speed analog
multiplier? Multiplier would help me. Something with carrier to 100MHz, or
200MHz, or 500MHz Harris (Intersil) seems to make a ADL???? 1.2GHz
multiplier, that looks like it has 60dB rejection up to 100MHz but that's
the dreaded typical spec, which means: who knows what?

 

On Thu, 17 Apr 2014 23:26:57 -0700, RobertMacy
<robert.a.macy@gmail.com> wrote:

On Thu, 17 Apr 2014 11:44:34 -0700, Simon S Aysdie <gwhite@ti.com> wrote:

...snip...
I still wonder why you can't filter them out. This is usually "easy,"
so this naturally leads to "why not?"


The carrier frequency is NOT fixed, and could go low enough that over 50
harmonics are lower than that first fundamental. Not easy to filter that
out. Tracking filters? can I get more thsn 60dB rejection to the third?
I'd like 80 dB. Any upper harmonic slipping in causes problems. Perhaps,
tracking filter will be equivalent. I like the idea of using hard driven
diode ring mixers for their simplicity. just have to add that elusive
tracking filter to rid of anything much above the fundamental. If this
were audio, I see a way to do it, but then again if audio, wouldn't have
to, multipliers exist.

Why do you insist of generate the modulated carrier at the final
frequency ?

Generate a fixed, say 300 MHz carrier, modulate it, run resolution
through a fixed 300 MHz filter with required bandwidth (typically
twice the modulating frequency). Mix this with a LO (VFO, PLL, DSS)
variable between 300.01 to 400 MHz, remove the image frequency
(600-700 MHz) with a simple LPF at 120 MHz.

Analog audio sweep generators for the 20 Hz to 20 kHz range in a
single sweep with linear sweep (xx Hz/s) has been used for at least 50
years. These generated a narrow band sweep at say 100-120 kHz and was
then mixed down to 0..20 kHz,followed by an LPF at say 30 kHz.

This is the opposite of how general coverage HF receivers (0..30 MHz)
have been operated for a few decades. The RF signal is mixed up to 45
MHz with the LO tuning between 45..75 MHz. At the 45 MHz IF the first
filter has a bandwidth of about 25 kHz, which is then mixed down to
10.7 MHz or 455 kHz for NBFM.

Spectrum analyzers usually work also in this upconverter mode, with
the first IF somewhere in 1..3 GHz.

 

[RobertMacy]

On Fri, 18 Apr 2014 00:56:37 -0700, <upsidedown@downunder.com> wrote:

...snip...

Why do you insist of generate the modulated carrier at the final
frequency ?

Generate a fixed, say 300 MHz carrier, modulate it, run resolution
through a fixed 300 MHz filter with required bandwidth (typically
twice the modulating frequency). Mix this with a LO (VFO, PLL, DSS)
variable between 300.01 to 400 MHz, remove the image frequency
(600-700 MHz) with a simple LPF at 120 MHz.

Analog audio sweep generators for the 20 Hz to 20 kHz range in a
single sweep with linear sweep (xx Hz/s) has been used for at least 50
years. These generated a narrow band sweep at say 100-120 kHz and was
then mixed down to 0..20 kHz,followed by an LPF at say 30 kHz.
This is the opposite of how general coverage HF receivers (0..30 MHz)
have been operated for a few decades. The RF signal is mixed up to 45
MHz with the LO tuning between 45..75 MHz. At the 45 MHz IF the first
filter has a bandwidth of about 25 kHz, which is then mixed down to
10.7 MHz or 455 kHz for NBFM.

Spectrum analyzers usually work also in this upconverter mode, with
the first IF somewhere in 1..3 GHz.

Thanks!!! Your description of the architecture jogged my mind enough to
envision a way to do this!

Now it's off to Minicircuits for parts.

 

On Fri, 18 Apr 2014 08:05:38 -0700, RobertMacy
<robert.a.macy@gmail.com> wrote:

On Fri, 18 Apr 2014 00:56:37 -0700, <upsidedown@downunder.com> wrote:

...snip...

Why do you insist of generate the modulated carrier at the final
frequency ?

Generate a fixed, say 300 MHz carrier, modulate it, run resolution
through a fixed 300 MHz filter with required bandwidth (typically
twice the modulating frequency). Mix this with a LO (VFO, PLL, DSS)
variable between 300.01 to 400 MHz, remove the image frequency
(600-700 MHz) with a simple LPF at 120 MHz.

Analog audio sweep generators for the 20 Hz to 20 kHz range in a
single sweep with linear sweep (xx Hz/s) has been used for at least 50
years. These generated a narrow band sweep at say 100-120 kHz and was
then mixed down to 0..20 kHz,followed by an LPF at say 30 kHz.
This is the opposite of how general coverage HF receivers (0..30 MHz)
have been operated for a few decades. The RF signal is mixed up to 45
MHz with the LO tuning between 45..75 MHz. At the 45 MHz IF the first
filter has a bandwidth of about 25 kHz, which is then mixed down to
10.7 MHz or 455 kHz for NBFM.

Spectrum analyzers usually work also in this upconverter mode, with
the first IF somewhere in 1..3 GHz.


Thanks!!! Your description of the architecture jogged my mind enough to
envision a way to do this!

Now it's off to Minicircuits for parts.

When using LO high side mixing, please remember to switch side band in
any SSB type transmission.

With LO high side mixing topologies, you also need to consider the
frequency accuracy and phase noise issues of the LO.

A 1 ppm frequency error at final carrier of 1 MHz is 1 Hz away.
However, in an upconverting system, in which the LO runs in 300-400
MHz range and in this case at 301 MHz, 1 ppm is 300 Hz frequency
error.

I have seen uncalibrated audio signal generators set to -100 Hz in
order to get a decent +20 Hz signal .

These days with NCO/DDS systemic running at several hundred MHz, this
should no longer be an issue.

 

[Simon S Aysdie]

On Thursday, April 17, 2014 11:26:57 PM UTC-7, Robert Macy wrote:

On Thu, 17 Apr 2014 11:44:34 -0700, Simon S Aysdie <gwhite@ti.com> wrote:

I still wonder why you can't filter them out. This is usually "easy,"
so this naturally leads to "why not?"

The carrier frequency is NOT fixed, and could go low enough that over 50
harmonics are lower than that first fundamental. Not easy to filter that
out. Tracking filters? can I get more thsn 60dB rejection to the third?
I'd like 80 dB. Any upper harmonic slipping in causes problems. Perhaps,
tracking filter will be equivalent. I like the idea of using hard driven
diode ring mixers for their simplicity. just have to add that elusive
tracking filter to rid of anything much above the fundamental. If this
were audio, I see a way to do it, but then again if audio, wouldn't have
to, multipliers exist.

Well, I think Mr Upside Down Under has given you the jist of the idea that I was more or less assuming (and that is why I didn't get it). Maybe that will work for you. Enjoy the spur charts!

> What do you work in? TI email address?

I am in T&M these days, and on the RX'er side. (Spec An, Signal monitoring)

The TI email address is long gone. I was there 2001 to late 2004, after Metricom's crash. I keep the google address because it is a dead end for spammers.

does TI make a high speed analog
multiplier? ...

Don't know. Don't think so. They spun off the WLAN stuff, which is where I was located.

 

[RobertMacy]

On Fri, 18 Apr 2014 11:36:20 -0700, <upsidedown@downunder.com> wrote:

...snip...

When using LO high side mixing, please remember to switch side band in
any SSB type transmission.

With LO high side mixing topologies, you also need to consider the
frequency accuracy and phase noise issues of the LO.

A 1 ppm frequency error at final carrier of 1 MHz is 1 Hz away.
However, in an upconverting system, in which the LO runs in 300-400
MHz range and in this case at 301 MHz, 1 ppm is 300 Hz frequency
error.

I have seen uncalibrated audio signal generators set to -100 Hz in
order to get a decent +20 Hz signal .

These days with NCO/DDS systemic running at several hundred MHz, this
should no longer be an issue.

After your excellent suggestion to become aggressive on the frequency
shifting, I blocked out the system into 'physically realizable' RF
components [I think} Once into that form, it was easy to see
'weaknesses', the sensitivity to phase noise, etc. So far, everything is
under 500MHz, which should make for easily obtainable components. One
thing that came out, was that the MOST sensitive transition can be made to
be fixed, which means it's possible to do a bit of tuning to enhance
performance.

Thanks again. My only defense for NOT pursuing this architecture earlier
is that I greatly feared getting into those noise sources - that's my
story, and I'm sticking to it.

 

[Phil Hobbs]

On 4/19/2014 12:52 PM, RobertMacy wrote:

On Fri, 18 Apr 2014 11:36:20 -0700, <upsidedown@downunder.com> wrote:

...snip...

When using LO high side mixing, please remember to switch side band in
any SSB type transmission.

With LO high side mixing topologies, you also need to consider the
frequency accuracy and phase noise issues of the LO.

A 1 ppm frequency error at final carrier of 1 MHz is 1 Hz away.
However, in an upconverting system, in which the LO runs in 300-400
MHz range and in this case at 301 MHz, 1 ppm is 300 Hz frequency
error.

I have seen uncalibrated audio signal generators set to -100 Hz in
order to get a decent +20 Hz signal .

These days with NCO/DDS systemic running at several hundred MHz, this
should no longer be an issue.


After your excellent suggestion to become aggressive on the frequency
shifting, I blocked out the system into 'physically realizable' RF
components [I think} Once into that form, it was easy to see
'weaknesses', the sensitivity to phase noise, etc. So far, everything is
under 500MHz, which should make for easily obtainable components. One
thing that came out, was that the MOST sensitive transition can be made
to be fixed, which means it's possible to do a bit of tuning to enhance
performance.

Thanks again. My only defense for NOT pursuing this architecture earlier
is that I greatly feared getting into those noise sources - that's my
story, and I'm sticking to it.

If you derive all the LOs from the same good quality crystal oscillator,
_and_keep_the_path_delays_the_same_, most of that added phase junk will
get subtracted again on the way down.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

 

On Sat, 19 Apr 2014 09:52:16 -0700, RobertMacy
<robert.a.macy@gmail.com> wrote:

On Fri, 18 Apr 2014 11:36:20 -0700, <upsidedown@downunder.com> wrote:

...snip...

When using LO high side mixing, please remember to switch side band in
any SSB type transmission.

With LO high side mixing topologies, you also need to consider the
frequency accuracy and phase noise issues of the LO.

A 1 ppm frequency error at final carrier of 1 MHz is 1 Hz away.
However, in an upconverting system, in which the LO runs in 300-400
MHz range and in this case at 301 MHz, 1 ppm is 300 Hz frequency
error.

I have seen uncalibrated audio signal generators set to -100 Hz in
order to get a decent +20 Hz signal .

These days with NCO/DDS systemic running at several hundred MHz, this
should no longer be an issue.


After your excellent suggestion to become aggressive on the frequency
shifting, I blocked out the system into 'physically realizable' RF
components [I think} Once into that form, it was easy to see
'weaknesses', the sensitivity to phase noise, etc. So far, everything is
under 500MHz, which should make for easily obtainable components. One
thing that came out, was that the MOST sensitive transition can be made to
be fixed, which means it's possible to do a bit of tuning to enhance
performance.

If you are using a DDS for frequency generation, some chips have a DDS
clock frequency premultiplier to allow use of fundamental mode
crystals (below 20 MHz) directly. These premultipliers are PLLs, which
might have an effect on phase noise, so check the specs with or
without premultiplier.

Since you apparently have only two frequencies f1 and f2 of interest,
you could even use a spreadsheet to validate your frequency
allocations and the locations of any spurs.

Put integer multiples (both positive and negative) horizontally and
multiples of f2 vertically on the spread sheet. In each cell,
calculate the sum of the column header and row header. In one cell,
there will be the wanted signal f1-f2 and in other cells various
unwanted frequencies.

To get a better visual overview, take the absolute value of the sum
(to reflect the negative sums) and compare the result the LPF
bandwidth, in your case somewhere between 120 to 150 MHz) and anything
above this can be filtered out and the cell should also be set to
spaces, instead of any numeric values, thus, any offending frequencies
will shown clearly.

Since the amplitude of the spurs seem to drop rapidly with higher
intermodulation products, you should not have to care e.g. for
-11 x f1 + 11 x f2 spurs (assuming you have tabulated from -11, ..,
0,, .+11 multiples both horizontally and vertically) this would be
the 22nd order spur and buried deep in thermal noise.

By varying f1 and/or f2 you might even find slightly lower values for
f1 and f2 with acceptable spur distribution.

 

[RobertMacy]

On Sat, 19 Apr 2014 10:44:22 -0700, Phil Hobbs <hobbs@electrooptical.net>
wrote:

...snip....
If you derive all the LOs from the same good quality crystal oscillator,
_and_keep_the_path_delays_the_same_, most of that added phase junk will
get subtracted again on the way down.

Cheers

Phil Hobbs

yes, COUNTING on that!