How to reach 1Mbaud in CAN with dsPIC33EP

[WalkOver]

Hello,

I use can-pic24_dspic33.c driver with a dsPIC33EP64GP502 MCU.
I was able to operate the bus at 125 250 and 500 kbps. But it is impossible to achieve 1Mbaud.

I tried at 50 100 and 140 Mhz using the internal RC oscillator with PLL.

The bus speed remains to 125 kbps with the error "CAN_EC_BAUD_INVALID".

How can I reach 1 Mbaud ? Is there a bug in the driver or do I need to use an external oscillator ?

Thank you for your help

[Ttelmah]

The faster you clock the bus, the lower the oscillator tolerance that is
allowed. At 1MB/sec, you are only allowed a tolerance of 0.49%. The
internal FRC oscillator only guarantees 0.9%....

Just try rebuilding, telling the compiler that you have a crystal oscillator.
If it accepts it, you have your answer.

[WalkOver]

Hello Ttelmah,

Thank you for your help.

I'm not sure I made myself clear.
The error I have is not from the compiler, it is handled by the driver and returned by the function can_set_baud()

I will try this tomorrow on real hardware. There may be a way to test this in simulation but I don't know enough CCS compiler yet for that.

[temtronic]

While I don't have that PIC or use CAN , anymore, I did look at the CCS example program code and saw this.....

[Ttelmah]

OK.
The first point is that even if you did set this rate up, it would probably
give issues using the Frc oscillator. It just is not accurate enough. See
at the end though.

The second though it to look at how can_set_baud works.
It counts through BRP values of 1 and upwards, calculating the bit time
in clocks using tBits = (Clock / (BRP * 2)) / Rate
It has to be able to get a division of the master clock that gives a result
for this that is between 8 and 25, or you will get the return BAUD_INVALID.
Now this also has to be met with the specified sample time (default 0.875
of the bit time).
What is happening is that with the specified sample point, and the clock
rates you are trying, it cannot find a divisor that works. Now the 0.875
value implies that there has to be a multiple of 8 clocks per bit. It may
simply be that the clock rate you are specifying does not give a divisor
meeting the parameters. So (for example), if I specify a clock of 100MHz,
this can't give a division by 8*1M, that will work. 100/8 = 12.5. Not an
integer division. 140MHz gives the same problem 140/8 = 17.5.
However selecting a CPU clock of 120MHz gives a legal division. 120/8
= 15.
So change your CPU speed to 120MHz, and the fault should disappear.
However be then aware that your accuracy will not be good enough
unless you add a crystal. Unless you have a very short bus. AN1798
from NXP gives the formulae for calculating the clock accuracies required
at different CAN rates. The actual accuracy needed, get tighter with
higher baud rates, and also with longer busses. The internal oscillator
is comfortably good enough for a 125K, 250K & 500K bus at reasonable
distances, but at 1Mbps, takes you out of tolerance for anything over a
very few meters...

[WalkOver]

[temtronic]

just a comment...
hmm, while it works 'on the bench' at 120MHz using the internal, it might fail in the 'real world'. The oscillator frequency will shift due to temperature variations, how much ?... you'll have to read the datasheet. Early PICs were bad, newer ones much better BUT at high baud rates, it doesn't take much of a shift to get 'out of spec'.

[Ttelmah]

Also the allowed tolerance drops with bus length. If you are working on
the bench with (say) a 2m bus, and then go to a real site with a 20m bus,
you may well find problems appear...

25/8 = 3.125

No wonder that didn't work.

Your allowable frequencies are
8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136MHz
Or you have to change the settings for the clock sampling point.

Jay, the tolerance on this Frc is very good indeed. Just +/- 0.9%. This is
good enough for reliable Canbus operation at the lower baud rates and
even over the full temperature range would probably work for the 1Mbps
bus for short busses. However it would fail for something like a full length
car bus at the higher rate for the temperatures these need....

[WalkOver]

Hello again,

I'm working on very short CAN bus and yes, this MCU has a more stable oscillator than other PIC family....but I bought a 2.5ppm TXO following your remarks and I have already made the modification on my PCB

I made a test in real wold with a 25 Mhz TXO but I bought a 16Mhz one to reach 120Mhz easily with the PLL.

[Ttelmah]

Good.
Nice when things move forwards.