Forums

watch dog on 56f807

Started by linelizabeth January 17, 2008
Our product is medium voltage (4160V) variable frequency drive. It
consists of 9~24 power cells depending on the power level of the
load. Each power cell has a control board with a DSP56F807 and an
Altera EPLD on it. The DSP talks to the EPLD via SPI port and the
EPLD talks to a master control board via a fiber optic cable.

The drives that we have DSP watch-dogged is used in a coal power
plant to control a motor which spins an ID fan. Over the past two
months, six power cells' DSPs have been watch-dogged.

In our lab, we have been trying to replicate the watch dog event
again on the cells that had been watch-dogged at the power plant and
sent back to us for diagnosis. We injected conductive and radiated RF
(up to over 300MHz and 110dB) to the cells, as well as high power
transient. The analog inputs to the DSP A/D were severely disturbed,
but so far we have not been able to get the DSP watch-dogged.

It is imperative that we find the root cause of the watch dog event
and come up with a viable solution for the customer. So I would
greatly appreciate it if someone can give some help.
I really need more information to try to help. How are the power cells connected to the DSP control board? A detailed block diagram would be helpful.

I assume "watch-dogged" means the processor is resetting due to watchdog time outs due to unknown causes.

linelizabeth wrote:
Our product is medium voltage (4160V) variable frequency drive. It
consists of 9~24 power cells depending on the power level of the
load. Each power cell has a control board with a DSP56F807 and an
Altera EPLD on it. The DSP talks to the EPLD via SPI port and the
EPLD talks to a master control board via a fiber optic cable.

The drives that we have DSP watch-dogged is used in a coal power
plant to control a motor which spins an ID fan. Over the past two
months, six power cells' DSPs have been watch-dogged.

In our lab, we have been trying to replicate the watch dog event
again on the cells that had been watch-dogged at the power plant and
sent back to us for diagnosis. We injected conductive and radiated RF
(up to over 300MHz and 110dB) to the cells, as well as high power
transient. The analog inputs to the DSP A/D were severely disturbed,
but so far we have not been able to get the DSP watch-dogged.

It is imperative that we find the root cause of the watch dog event
and come up with a viable solution for the customer. So I would
greatly appreciate it if someone can give some help.
The given informations are not sufficient to find the problem.
Anyway, in general, I can suggest you to be careful with variables
shared from interrupts and the other code.
When the normal execution code needs to write to a variable that can be
written by an interrupt, too, you have two ways to safely do it:
use an atomic write operation, or disable interrupts before writing the
variable. If you choose the second one, be sure that the pipeline is
exhausted before writing the variable (chapter 7 of DSP56800 Family
Manual).
The same thing if you have nested interrupts that share variables.

If you don't follow these guidelines, you could have watchdog or reset
events.

Regards,
Roberto

-----Messaggio originale-----
Da: m... [mailto:m...] Per
conto di linelizabeth
Inviato: mercoled16 gennaio 2008 17.19
A: m...
Oggetto: [motoroladsp] watch dog on 56f807

Our product is medium voltage (4160V) variable frequency drive. It
consists of 9~24 power cells depending on the power level of the
load. Each power cell has a control board with a DSP56F807 and an
Altera EPLD on it. The DSP talks to the EPLD via SPI port and the
EPLD talks to a master control board via a fiber optic cable.

The drives that we have DSP watch-dogged is used in a coal power
plant to control a motor which spins an ID fan. Over the past two
months, six power cells' DSPs have been watch-dogged.

In our lab, we have been trying to replicate the watch dog event
again on the cells that had been watch-dogged at the power plant and
sent back to us for diagnosis. We injected conductive and radiated RF
(up to over 300MHz and 110dB) to the cells, as well as high power
transient. The analog inputs to the DSP A/D were severely disturbed,
but so far we have not been able to get the DSP watch-dogged.

It is imperative that we find the root cause of the watch dog event
and come up with a viable solution for the customer. So I would
greatly appreciate it if someone can give some help.