Prev: Circuit Design Using PADS Layout
Next: USB spectrum analyzer
From: Nemo on 3 Oct 2009 09:50
I'd appreciate some witty put-downs and scathing feedback on the
following.

* I want to take the opportunity of a circuit redesign on a piece of
scientific instrumentation to use a differential output to reduce EMC
problems, but our industry normally uses single-ended analog signals for
historical reasons, ie they tie one input line to their circuit 0V. Does
this make use of differential signals pointless? I'm using an isolated
supply to power my circuit, so it's floating and will break any ground
loops - I hope. I was thinking of converting my own single-ended signal
to differential down the connecting cable using an ADA4922-1 from Analog
Devices, which does the conversion simply.

* Examples I've seen show differential-drive outputs used with twisted
pair. Is there any reason not to use coax? My highest signal freq is
500kHz, so I'm thinking TP would actually be better as I understand coax
is really only effective at screening higher freqs.

* Our industry uses 50 ohm coax. The differential amplifier datasheets
I've scanned imply they are designed for loads of 75 or 100 ohms. Are
differential amps so specialised for driving ADC's that they aren't
suitable for driving 50 ohm coax?

* I only have to transmit analog signals of about 500kHz bandwith over 4
metres of cable. I strongly suspect that at least one of the bits of
equipment I'm expected to drive has an input impedance of about 500
ohms, ie is poorly matched. It doesn't seem to have any reflection
problems with the bit of kit I'm replacing, which used an AD8045 video
line driver, so I guess reflections etc aren't much of an issue. Am I
being naive here - for example would you expect problems driving a
poorly terminated line with a diff amp rated at 100'sMHz GBW driving it?

* A possible circuit gotcha is, I want to amplify down to near DC (say
10Hz) and retain the DC offset information in the output so transformers
are ruled out. I don't care if the offset is shifted as long as it's
constant.

* What are folks' favourite ICs for single-ended to differential output
conversion?

Thank you,
--
Nemo
From: John Larkin on 3 Oct 2009 11:31
On Sat, 3 Oct 2009 14:50:15 +0100, Nemo
<zzz(a)nospam.nospam.nospam.nospam.co.uk> wrote:

>I'd appreciate some witty put-downs and scathing feedback on the
>following.
>
>* I want to take the opportunity of a circuit redesign on a piece of
>scientific instrumentation to use a differential output to reduce EMC
>problems, but our industry normally uses single-ended analog signals for
>historical reasons, ie they tie one input line to their circuit 0V. Does
>this make use of differential signals pointless? I'm using an isolated
>supply to power my circuit, so it's floating and will break any ground
>loops - I hope. I was thinking of converting my own single-ended signal
>to differential down the connecting cable using an ADA4922-1 from Analog
>Devices, which does the conversion simply.
>
>* Examples I've seen show differential-drive outputs used with twisted
>pair. Is there any reason not to use coax? My highest signal freq is
>500kHz, so I'm thinking TP would actually be better as I understand coax
>is really only effective at screening higher freqs.
>
>* Our industry uses 50 ohm coax. The differential amplifier datasheets
>I've scanned imply they are designed for loads of 75 or 100 ohms. Are
>differential amps so specialised for driving ADC's that they aren't
>suitable for driving 50 ohm coax?
>
>* I only have to transmit analog signals of about 500kHz bandwith over 4
>metres of cable. I strongly suspect that at least one of the bits of
>equipment I'm expected to drive has an input impedance of about 500
>ohms, ie is poorly matched. It doesn't seem to have any reflection
>problems with the bit of kit I'm replacing, which used an AD8045 video
>line driver, so I guess reflections etc aren't much of an issue. Am I
>being naive here - for example would you expect problems driving a
>poorly terminated line with a diff amp rated at 100'sMHz GBW driving it?
>
>* A possible circuit gotcha is, I want to amplify down to near DC (say
>10Hz) and retain the DC offset information in the output so transformers
>are ruled out. I don't care if the offset is shifted as long as it's
>constant.
>
>* What are folks' favourite ICs for single-ended to differential output
>conversion?

At these speeds, just two opamps, one inverting and one non-inverting.
Something fast to keep the phases balanced.

The ADA4922-1 pretty much does this. Nice part, lots of swing and lots
of power dissipation. Drive two coaxes with precision 50-ohm source
terminator resistors. Ground the shields on both ends. Source
termination reduces the dynamic load on the opamps and helps keep
distortion and power dissipation down. Don't terminate the receive
end.

What's your dynamic range? What's the physics? I'm doing a very
similar thing, moving a wide-dynamic-range signal across a room on two
coaxes, KHz to 20 MHz, and need to reject ambient RF. We're going for
maximum swing (20 v p-p differential) to climb above the noise.
Source-only termination gives 1.00000 gain through each cable leg,
maintaining the CMRR.

John

From: MooseFET on 3 Oct 2009 12:49
On Oct 3, 8:31 am, John Larkin
<jjlar...(a)highNOTlandTHIStechnologyPART.com> wrote:
> On Sat, 3 Oct 2009 14:50:15 +0100, Nemo
>
>
>
> <z...(a)nospam.nospam.nospam.nospam.co.uk> wrote:
> >I'd appreciate some witty put-downs and scathing feedback on the
> >following.
>
> >* I want to take the opportunity of a circuit redesign on a piece of
> >scientific instrumentation to use a differential output to reduce EMC
[....]
> The ADA4922-1 pretty much does this. Nice part, lots of swing and lots
> of power dissipation. Drive two coaxes with precision 50-ohm source
> terminator resistors.

Make or buy some of that stuff with 2 coax cables glued side by side.
You want to keep the enclosed area between the shields small. Even
the best coax is not 100% shielded. This will tend to make a current
flow in the shields. It isn't a problem unless the area enclosed is
very large.

> Ground the shields on both ends. Source
> termination reduces the dynamic load on the opamps and helps keep
> distortion and power dissipation down. Don't terminate the receive
> end.

Keep the capacitance small at the receive end. You want to go from Z
= 50 ohms to infinite. Short PCB traces will do this.

If you need suppress EMI at the transmit end, run the whole mess
through a single ring core

> What's your dynamic range? What's the physics? I'm doing a very
> similar thing, moving a wide-dynamic-range signal across a room on two
> coaxes, KHz to 20 MHz, and need to reject ambient RF. We're going for
> maximum swing (20 v p-p differential) to climb above the noise.
> Source-only termination gives 1.00000 gain through each cable leg,
> maintaining the CMRR.
>
> John

From: Nemo on 3 Oct 2009 13:27
In article <qiqec5hifohvv3qhq5f25b6keq11g7lnih(a)4ax.com>, John Larkin
<jjlarkin(a)highNOTlandTHIStechnologyPART.com> writes
>On Sat, 3 Oct 2009 14:50:15 +0100, Nemo
><zzz(a)nospam.nospam.nospam.nospam.co.uk> wrote:
>
>>I'd appreciate some witty put-downs and scathing feedback on the
>>following.
>>
>>* I want to take the opportunity of a circuit redesign on a piece of
>>scientific instrumentation to use a differential output to reduce EMC
>>problems, but our industry normally uses single-ended analog signals for
>>historical reasons, ie they tie one input line to their circuit 0V. Does
>>this make use of differential signals pointless? I'm using an isolated
>>supply to power my circuit, so it's floating and will break any ground
>>loops - I hope. I was thinking of converting my own single-ended signal
>>to differential down the connecting cable using an ADA4922-1 from Analog
>>Devices, which does the conversion simply.
>>
>>* Examples I've seen show differential-drive outputs used with twisted
>>pair. Is there any reason not to use coax? My highest signal freq is
>>500kHz, so I'm thinking TP would actually be better as I understand coax
>>is really only effective at screening higher freqs.
>>
>>* Our industry uses 50 ohm coax. The differential amplifier datasheets
>>I've scanned imply they are designed for loads of 75 or 100 ohms. Are
>>differential amps so specialised for driving ADC's that they aren't
>>suitable for driving 50 ohm coax?
>>
>>* I only have to transmit analog signals of about 500kHz bandwith over 4
>>metres of cable. I strongly suspect that at least one of the bits of
>>equipment I'm expected to drive has an input impedance of about 500
>>ohms, ie is poorly matched. It doesn't seem to have any reflection
>>problems with the bit of kit I'm replacing, which used an AD8045 video
>>line driver, so I guess reflections etc aren't much of an issue. Am I
>>being naive here - for example would you expect problems driving a
>>poorly terminated line with a diff amp rated at 100'sMHz GBW driving it?
>>
>>* A possible circuit gotcha is, I want to amplify down to near DC (say
>>10Hz) and retain the DC offset information in the output so transformers
>>are ruled out. I don't care if the offset is shifted as long as it's
>>constant.
>>
>>* What are folks' favourite ICs for single-ended to differential output
>>conversion?
>
>At these speeds, just two opamps, one inverting and one non-inverting.
>Something fast to keep the phases balanced.
>
>The ADA4922-1 pretty much does this. Nice part, lots of swing and lots
>of power dissipation. Drive two coaxes with precision 50-ohm source
>terminator resistors. Ground the shields on both ends. Source
>termination reduces the dynamic load on the opamps and helps keep
>distortion and power dissipation down. Don't terminate the receive
>end.
>
>What's your dynamic range? What's the physics? I'm doing a very
>similar thing, moving a wide-dynamic-range signal across a room on two
>coaxes, KHz to 20 MHz, and need to reject ambient RF. We're going for
>maximum swing (20 v p-p differential) to climb above the noise.
>Source-only termination gives 1.00000 gain through each cable leg,
>maintaining the CMRR.

Thanks John. To answer your questions,

I am trying to improve a photodiode amplifier. Dynamic range is about 16
bits, ie the peak signal coming out of the amplification section is up
to 4V peak (always positive-going above 0V), this is presently fed down
a cable to a 16 bit ADC in another box. This other box, which I can't
change, would be unhappy with inputs over about 5V. I believe the
circuit is likely to be re-used for other similar applications with
other equipment, which would typically expect inputs up to +/-5V so
should be happy with this 0 to 4V signal.

Starting point of the circuit improvement is to convert the extant
preamp to Phil Hobbs' cascode circuit which I believe will reduce noise
/ increase bandwidth. I thought I may as well try improving the circuit
in other ways, to get experience with new techniques we could use on a
range of products. I've considered and dismissed a couple of tricks I
figured were too high risk like a notch filter to get rid of hum (this
would give phase shift as you swept across its notch freq, and some
people look at signals right down to 20Hz); but I've added a 300kHz
linear phase filter to reduce HF fuzz, and an isolated power supply to
get rid of earth loops.

Concerning the earth loops: this market - electron microscopy - is full
of weird metal-boxed instrumentation from different manufacturers which
people buy and hook together with coax, without thinking about earth
loops. This is usually 50 ohm coax terminated with BNC's. So, I'm not
sure I can use two separate coax's for my output as I may have problems
connecting it to legacy equipment - but I'll have a good think about
this. I like that idea.

The monitoring equipment is placed a few metres away from the 'scopes as
they are very sensitive to ambient magnetic fields. This equipment is
several other boxes ending in a PC to clean up and display a picture of
the sample being peered at. A typical cable run from the 'scope is 3 to
5 metres, under floorboards so you don't trip over them, plus a whole
mess of interconnecting coax cables behind the PC and other boxes. This
leads to ground loops when people plug things into different mains
sockets around the lab... and just a small ground current in these loops
can physically wobble the electron beam in the scope, or create enough
interference in the high gain amplifiers, to cause noticeable banding at
60Hz.

Users are interested in the DC level of the signal (corresponding to
photodiode dark current of a few nA) as well as AC signals down to
around 20Hz. Highest frequency of interest is about 50kHz, but it turns
out the people specifying this aren't really very clear what they mean
by bandwidth, thus my low pass filter is set to 300kHz to give
reasonably sharp risetimes for sharp-edged features.

Although Beta users found no EMC problems except ground loops with the
predecessor product, I had trouble getting it through formal EMC testing
due largely to poor rejection of radiated RFI pickup in the preamp, and
conducted ripple getting in to the test system as a whole (several boxes
joined by cables, but I'm not redesigning most of them) through the
mains. Because the system was designed one box at a time, by different
folk, the overall earth loops weren't properly thought out, but the weak
point was the preamp I'm working on because small problems there were
amplified. The emissions from the system were not, in general, a
problem. So my personal target is to not just improve signal-to-noise
ratio and bandwidth, but make the thing bulletproof to EMC. There are
some constraints from what I have to interface to, but I have a lot of
freedom as to how I implement the box I'm working on.
--
Nemo
From: Bill Sloman on 4 Oct 2009 10:39
On Oct 3, 7:27 pm, Nemo <z...(a)nospam.nospam.nospam.nospam.co.uk>
wrote:
> In article <qiqec5hifohvv3qhq5f25b6keq11g7l...(a)4ax.com>, John Larkin
> <jjlar...(a)highNOTlandTHIStechnologyPART.com> writes
>
> >On Sat, 3 Oct 2009 14:50:15 +0100, Nemo
> ><z...(a)nospam.nospam.nospam.nospam.co.uk> wrote:
>
> >>I'd appreciate some witty put-downs and scathing feedback on the
> >>following.
>
> >>* I want to take the opportunity of a circuit redesign on a piece of
> >>scientific instrumentation to use a differential output to reduce EMC
> >>problems, but our industry normally uses single-ended analog signals for
> >>historical reasons, ie they tie one input line to their circuit 0V. Does
> >>this make use of differential signals pointless? I'm using an isolated
> >>supply to power my circuit, so it's floating and will break any ground
> >>loops - I hope. I was thinking of converting my own single-ended signal
> >>to differential down the connecting cable using an ADA4922-1 from Analog
> >>Devices, which does the conversion simply.
>
> >>* Examples I've seen show differential-drive outputs used with twisted
> >>pair. Is there any reason not to use coax? My highest signal freq is
> >>500kHz, so I'm thinking TP would actually be better as I understand coax
> >>is really only effective at screening higher freqs.
>
> >>* Our industry uses 50 ohm coax. The differential amplifier datasheets
> >>I've scanned imply they are designed for loads of 75 or 100 ohms. Are
> >>differential amps so specialised for driving ADC's that they aren't
> >>suitable for driving 50 ohm coax?
>
> >>* I only have to transmit analog signals of about 500kHz bandwith over 4
> >>metres of cable. I strongly suspect that at least one of the bits of
> >>equipment I'm expected to drive has an input impedance of about 500
> >>ohms, ie is poorly matched. It doesn't seem to have any reflection
> >>problems with the bit of kit I'm replacing, which used an AD8045 video
> >>line driver, so I guess reflections etc aren't much of an issue. Am I
> >>being naive here - for example would you expect problems driving a
> >>poorly terminated line with a diff amp rated at 100'sMHz GBW driving it?
>
> >>* A possible circuit gotcha is, I want to amplify down to near DC (say
> >>10Hz) and retain the DC offset information in the output so transformers
> >>are ruled out. I don't care if the offset is shifted as long as it's
> >>constant.
>
> >>* What are folks' favourite ICs for single-ended to differential output
> >>conversion?
>
> >At these speeds, just two opamps, one inverting and one non-inverting.
> >Something fast to keep the phases balanced.
>
> >The ADA4922-1 pretty much does this. Nice part, lots of swing and lots
> >of power dissipation. Drive two coaxes with precision 50-ohm source
> >terminator resistors. Ground the shields on both ends. Source
> >termination reduces the dynamic load on the opamps and helps keep
> >distortion and power dissipation down. Don't terminate the receive
> >end.
>
> >What's your dynamic range? What's the physics? I'm doing a very
> >similar thing, moving a wide-dynamic-range signal across a room on two
> >coaxes, KHz to 20 MHz, and need to reject ambient RF. We're going for
> >maximum swing (20 v p-p differential) to climb above the noise.
> >Source-only termination gives 1.00000 gain through each cable leg,
> >maintaining the CMRR.
>
> Thanks John. To answer your questions,
>
> I am trying to improve a photodiode amplifier. Dynamic range is about 16
> bits, ie the peak signal coming out of the amplification section is up
> to 4V peak (always positive-going above 0V), this is presently fed down
> a cable to a 16 bit ADC in another box. This other box, which I can't
> change, would be unhappy with inputs over about 5V. I believe the
> circuit is likely to be re-used for other similar applications with
> other equipment, which would typically expect inputs up to +/-5V so
> should be happy with this 0 to 4V signal.
>
> Starting point of the circuit improvement is to convert the extant
> preamp to Phil Hobbs' cascode circuit which I believe will reduce noise
> / increase bandwidth. I thought I may as well try improving the circuit
> in other ways, to get experience with new techniques we could use on a
> range of products. I've considered and dismissed a couple of tricks I
> figured were too high risk like a notch filter to get rid of hum (this
> would give phase shift as you swept across its notch freq, and some
> people look at signals right down to 20Hz); but I've added a 300kHz
> linear phase filter to reduce HF fuzz, and an isolated power supply to
> get rid of earth loops.
>
> Concerning the earth loops: this market - electron microscopy - is full
> of weird metal-boxed instrumentation from different manufacturers which
> people buy and hook together with coax, without thinking about earth
> loops. This is usually 50 ohm coax terminated with BNC's. So, I'm not
> sure I can use two separate coax's for my output as I may have problems
> connecting it to legacy equipment - but I'll have a good think about
> this. I like that idea.
>
> The monitoring equipment is placed a few metres away from the 'scopes as
> they are very sensitive to ambient magnetic fields. This equipment is
> several other boxes ending in a PC to clean up and display a picture of
> the sample being peered at. A typical cable run from the 'scope is 3 to
> 5 metres, under floorboards so you don't trip over them, plus a whole
> mess of interconnecting coax cables behind the PC and other boxes. This
> leads to ground loops when people plug things into different mains
> sockets around the lab... and just a small ground current in these loops
> can physically wobble the electron beam in the scope, or create enough
> interference in the high gain amplifiers, to cause noticeable banding at
> 60Hz.
>
> Users are interested in the DC level of the signal (corresponding to
> photodiode dark current of a few nA) as well as AC signals down to
> around 20Hz. Highest frequency of interest is about 50kHz, but it turns
> out the people specifying this aren't really very clear what they mean
> by bandwidth, thus my low pass filter is set to 300kHz to give
> reasonably sharp risetimes for sharp-edged features.
>
> Although Beta users found no EMC problems except ground loops with the
> predecessor product, I had trouble getting it through formal EMC testing
> due largely to poor rejection of radiated RFI pickup in the preamp, and
> conducted ripple getting in to the test system as a whole (several boxes
> joined by cables, but I'm not redesigning most of them) through the
> mains. Because the system was designed one box at a time, by different
> folk, the overall earth loops weren't properly thought out, but the weak
> point was the preamp I'm working on because small problems there were
> amplified. The emissions from the system were not, in general, a
> problem. So my personal target is to not just improve signal-to-noise
> ratio and bandwidth, but make the thing bulletproof to EMC. There are
> some constraints from what I have to interface to, but I have a lot of
> freedom as to how I implement the box I'm working on.

I had just this kind of problem with stuff I was designing for my
fancy electron-microscope-based electron beam tester back around 1990.

We put together a subtracting (current-feedback) amplifier with a
balun to ship a video signal around the machine without creating
grounds loops. It seemed to work. I've got the circuit somewhere - e-
mail me (bill.sloman(a)ieee.org will work) and I'll send you a scanned
image of that part of the circuit,

We weren't too fussed about the DC accuracy at the time, but you'd
have to find a currently available op amp for the job anyway.

--
Bill Sloman, Nijmegen
 | 
Pages: 1
Prev: Circuit Design Using PADS Layout
Next: USB spectrum analyzer