ISA Bus termination question

[Matt]

We were having an issue with a system that uses an ISA bus locking up
once every few hours. After a lot of work, we narrowed the issue down
to bus termination or lack thereof. After throwing some terminator
resistors on the ISA Bus the system runs fine for weeks. Great,
problem solved. My current task is figuring out how having no
termination caused the system to lock up after a few hours. I started
probing the clock, address lines, and data lines on the bus with and
without the termination. When there is no termination, there is a
larger overshoot and undershoot on the rising and falling edge
respectively however the delta of the termination vs. no termination
is less than a volt. I was hoping to see something a bit more drastic
but I am unsure what the threshold is on the overshoot/undershoot
voltages. Also, for grins, I did a FFT on the BCLK signal (8.333MHz)
and I noticed that the unterminated signal had well defined even
harmonics. The terminated signal had some noise around the even
harmonics but they were no where near as structured. I should also
point out the the odd harmonics on the terminated signal had a more
narrow band.

There are obvious differences in the terminated and unterminated
signals, but nothing that really stands out as the problem causing the
issue. Is there something else I should look at?

Thank you

 

[Jan Panteltje]

There are obvious differences in the terminated and unterminated
signals, but nothing that really stands out as the problem causing the
issue. Is there something else I should look at?

Thank you

Last time I did an ISA card it used no termination IIRC.
But that was a LONG time ago....

Capacitive load, crosstalk, DMA signal (can't remember what it was called,
but should be in decode).

 

[Jan Panteltje]

Capacitive load, crosstalk, DMA signal (can't remember what it was called,
but should be in decode).
Oh yes, remember now:
AEN (address enable), says no DMA cycle in progress, so valid address.

 

[Baron]

We were having an issue with a system that uses an ISA bus locking up
once every few hours. After a lot of work, we narrowed the issue down
to bus termination or lack thereof. After throwing some terminator
resistors on the ISA Bus the system runs fine for weeks. Great,
problem solved. My current task is figuring out how having no
termination caused the system to lock up after a few hours. I started
probing the clock, address lines, and data lines on the bus with and
without the termination. When there is no termination, there is a
larger overshoot and undershoot on the rising and falling edge
respectively however the delta of the termination vs. no termination
is less than a volt. I was hoping to see something a bit more drastic
but I am unsure what the threshold is on the overshoot/undershoot
voltages. Also, for grins, I did a FFT on the BCLK signal (8.333MHz)
and I noticed that the unterminated signal had well defined even
harmonics. The terminated signal had some noise around the even
harmonics but they were no where near as structured. I should also
point out the the odd harmonics on the terminated signal had a more
narrow band.

There are obvious differences in the terminated and unterminated
signals, but nothing that really stands out as the problem causing the
issue. Is there something else I should look at?

Thank you

Just a WAG ! Standing waves and jitter. The jitter moves the signal
edges about, at some point the standing wave pattern cancels out.
result missed edge.

 

[tlbs101]

We were having an issue with a system that uses an ISA bus locking up
once every few hours. After a lot of work, we narrowed the issue down
to bus termination or lack thereof. After throwing some terminator
resistors on the ISA Bus the system runs fine for weeks. Great,
problem solved. My current task is figuring out how having no
termination caused the system to lock up after a few hours. I started
probing the clock, address lines, and data lines on the bus with and
without the termination. When there is no termination, there is a
larger overshoot and undershoot on the rising and falling edge
respectively however the delta of the termination vs. no termination
is less than a volt. I was hoping to see something a bit more drastic
but I am unsure what the threshold is on the overshoot/undershoot
voltages. Also, for grins, I did a FFT on the BCLK signal (8.333MHz)
and I noticed that the unterminated signal had well defined even
harmonics. The terminated signal had some noise around the even
harmonics but they were no where near as structured. I should also
point out the the odd harmonics on the terminated signal had a more
narrow band.

There are obvious differences in the terminated and unterminated
signals, but nothing that really stands out as the problem causing the
issue. Is there something else I should look at?

Thank you

I worked on a PC-104 (ISA) project 2 years ago. We used a commercial
PC. I developed one board (various interfaces and video) while a
fellow engineer developed another board (with a 2-channel servo motor
controller). His was 16-bit ISA and mine was 8-bit. Both used FPGAs
to decode the ISA control signals. One problem we had was noise from
his board getting on the reset line quite often -- resetting the
system every few minutes. A simple RC filter on the reset line fixed
that problem.

Another problem we had was my 8-bit board was (initially) ignoring the
BALE signal. I had to re-write my FPGA code to accomondate the 16-bit
signals (including BALE), even though my board was 8-bit.

Hopefully one of those 2 bits of info help.

Tom P.

 

[David L. Jones]

We were having an issue with a system that uses an ISA bus locking up
once every few hours. After a lot of work, we narrowed the issue down
to bus termination or lack thereof. After throwing some terminator
resistors on the ISA Bus the system runs fine for weeks. Great,
problem solved. My current task is figuring out how having no
termination caused the system to lock up after a few hours. I started
probing the clock, address lines, and data lines on the bus with and
without the termination. When there is no termination, there is a
larger overshoot and undershoot on the rising and falling edge
respectively however the delta of the termination vs. no termination
is less than a volt. I was hoping to see something a bit more drastic
but I am unsure what the threshold is on the overshoot/undershoot
voltages. Also, for grins, I did a FFT on the BCLK signal (8.333MHz)
and I noticed that the unterminated signal had well defined even
harmonics. The terminated signal had some noise around the even
harmonics but they were no where near as structured. I should also
point out the the odd harmonics on the terminated signal had a more
narrow band.

There are obvious differences in the terminated and unterminated
signals, but nothing that really stands out as the problem causing the
issue. Is there something else I should look at?

Thank you

How exactly did you so confidently narrow down the problem to bus
termination?
Just because adding bus termination appears to fix the problem does
not mean that lack of termination was the only issue, as you are now
likely finding out buy having to explain it the exact cause.
It is highly likely there is some other issue that the termination
just happens to be masking. You might find your device fails again
over temperature or some other variable perhaps.

Normally an ISA bus does not need termination, but the option is
there.
This comes from the PC-104 standard:

"4.7 Bus Termination Option
Termination of the 8-bit and 16-bit ISA bus signals may be desired in
some systems to increase
data integrity and system reliability. When termination is included,
AC termination networks
must be used to provide termination close to the characteristic
impedance of the signal lines
without exceeding the DC output current capabilities of the drivers.
As in the ISA standard, the recommended network consists of a resistor-
capacitor network of 40-
60 ohms in series with 30-70 pF, connected between each bus signal and
ground.
Whether termination is needed, and where it should be located, is
dependent on the specific
system configuration and must be determined by the system designer."

Also, how did you probe the signals?
The measured "bounce" on the "terminated" signal is dependent entirely
upon the ground point at which you measure it. Did you actually use
proper low inductive probing techniques at every receiver chip on the
bus? I'm immediately sceptical when someone says they have measured
and compared overshoot and undershoot.

Poor grounding and/or power decoupling layout is a more likely cause
than simply not having the bus terminated.

Dave.

 

[Matt]

How exactly did you so confidently narrow down the problem to bus
termination?
Just because adding bus termination appears to fix the problem does
not mean that lack of termination was the only issue, as you are now
likely finding out buy having to explain it the exact cause.
It is highly likely there is some other issue that the termination
just happens to be masking. You might find your device fails again
over temperature or some other variable perhaps.

Normally an ISA bus does not need termination, but the option is
there.
This comes from the PC-104 standard:

"4.7 Bus Termination Option
Termination of the 8-bit and 16-bit ISA bus signals may be desired in
some systems to increase
data integrity and system reliability. When termination is included,
AC termination networks
must be used to provide termination close to the characteristic
impedance of the signal lines
without exceeding the DC output current capabilities of the drivers.
As in the ISA standard, the recommended network consists of a resistor-
capacitor network of 40-
60 ohms in series with 30-70 pF, connected between each bus signal and
ground.
Whether termination is needed, and where it should be located, is
dependent on the specific
system configuration and must be determined by the system designer."

Also, how did you probe the signals?
The measured "bounce" on the "terminated" signal is dependent entirely
upon the ground point at which you measure it. Did you actually use
proper low inductive probing techniques at every receiver chip on the
bus? I'm immediately sceptical when someone says they have measured
and compared overshoot and undershoot.

Poor grounding and/or power decoupling layout is a more likely cause
than simply not having the bus terminated.

Dave.

Thank you all for the responses.

Dave, what you say is true. It's very possible that termination may
have only delayed the failure so there is quite a bit more testing to
be done. The reference I used for the measurments was the actual
ground on the motherboard's bus driver ICs and I am using a low
impedance/capacitance probe. I will let you all know if I find
anything else that stands out.

Thank you again,
Matt

 

[rickman]

Thank you all for the responses.

Dave, what you say is true. It's very possible that termination may
have only delayed the failure so there is quite a bit more testing to
be done. The reference I used for the measurments was the actual
ground on the motherboard's bus driver ICs and I am using a low
impedance/capacitance probe. I will let you all know if I find
anything else that stands out.

Thank you again,
Matt

One thing that a lot of people forget is that the clk signal on ISA is
not really a timing signal for the bus. It is typically used by chips
that interface to the bus, but not as a synchronous clock. Instead,
the read and write strobes are the timing signals that should be
treated as clocks. Depending on how a given chip detects the edges of
these "clocks" an over-shoot or under-shoot with ringing can result in
double clocking. So rather than looking at the bus clock, have you
looked carefully at the read and write strobes? Otherwise, I would
suspect the timing rather than the signal integrity on the bus.
Certainly ringing on the data and address lines is not significant
unless it is so severe that it causes latchup in the chips or delays
the settling time enough to cause a timing error.

The terminator likely created some timing differences. If you want to
explore the matter you can try removing terminators until you find a
particular one that can cause or fix the problem. With the problem
taking hours to occur, this may be a difficult problem to isolate.
But once you have, then you can explore the timing changes and see
exactly what is happening.

Oh, like David said, your ground for a ringing measurement is *very*
important. You need to pick a ground **very** close to the pin on the
receiver chip and use a ***very*** short ground lead. If you remove
the clip on the end of the scope probe and wrap a paper clip around
the ground sleeve, you can get the probe and the clip to contact the
board at the same time with a steady hand. This will give you an
accurate picture of the ringing on each receiver.

Rick

 

[David L. Jones]

Thank you all for the responses.

Dave, what you say is true. It's very possible that termination may
have only delayed the failure so there is quite a bit more testing to
be done. The reference I used for the measurments was the actual
ground on the motherboard's bus driver ICs and I am using a low
impedance/capacitance probe. I will let you all know if I find
anything else that stands out.

Thank you again,
Matt

Good luck with it, sounds like it'll be a tricky one to find, but
probably obvious with hindsight!

Dave.