problems with high baud rates and #use rs232

[jroth]

my program uses a timer1 interrupt to run PWM on a number of pins of a PIC 16F877. i listen over serial (#use rs232) to commands from a Rabbit RCM2200 microcontroller. the program works fine at 28.8k baud rate, but i am running into problems when i try to increase the baud rate to 57,600. i am clocking the PIC at 20 MHz.

i'm not sure how #use RS232 configures the USART, and perhaps the problem is linked to this ... maybe a buffer gets overloaded with too much incomming data.

another problem could be the interrupt frequency ... but i tried slowing this to 1/10 of the current speed, and it still can't handle 57.6k.

the program will run for a few seconds at 57.6k, but then crashes - it won't communicate over serial. i am sending data at the rate of 30 characters a second, so i'm surprised by the results.

here is the source:

[PCM programmer]

Well, your isr is massively long. Maybe you can't help that.
But at 57600 baud, one character period is 173 us.
The hardware USART's receive fifo is 2-deep, plus the
incoming shift register.

I didn't figure out exactly how long your ISR takes,
I just glanced at it. But it's got to go through at least
one of those 30-iteration for-loops every interrupt.
The amount of ASM code is massive, including the
calls. It adds up.

I think you need to time it. Either do it empirically
with logic analyzer or scope (put in code to put out
a pulse at the start of the ISR, and another pulse at
the end), or add it up by hand. That analysis will
tell you if it's too long and is causing you to lose chars.

[Sergio]

I am using a software interrupt at 115200 baud (20Mhz xtal) without any problems. I did discover that kbhit was too slow for those speeds. Instead I rolled up a simpler one that just keeps looking if the line at the pin went down. I am making the assumption that if it goes down then a character is being sent.

I also never recive more than two characters at a time hence a buffer of only 2. Which makes buffering simpler.

I need to simplify it but here's my code:

#int_ext
void serial2()
{
Display_LCD_Flag=1;
timeout=0;
timeout_error=FALSE;

#use RS232(baud=115200, xmit=PIN_B6, rcv=PIN_B0) //Amulet Software port
while(RCV_PIN&&(++timeout<765)); //This gives about 2ms timeout

if(!RCV_PIN)
serial_rcv[0]=getc();
else
timeout_error=TRUE;

timeout=0;
while(RCV_PIN&&(++timeout<765)); //This gives about 2ms timeout

if(!RCV_PIN)
serial_rcv[1]=getc();
else
timeout_error=TRUE;

menu=serial_rcv[1];
}
_________________
Sergio

[ckielstra]

As PCM Programmer already said: 57000 baud results in 173us per character.
At 20MHz this is equivalent to 865 instructions. Your interrupt routine should take no longer to execute than this.

A quick study of the assembly code of your interrupt handler showed you are using a minimum of 420 instructions in every interrupt. This is safe, but every 100th interrupt you are using at least 1320 instructions and with all leds on this can go up to at least 2300 instructions. This is too long!

The UART has a 2 bytes FIFO buffer, but your interrupt routine takes too long to execute and you will get an occasional overrun error. This error will be automatically cleared by the 'errors' parameter in your #use rs232 statement. Your protocol will be 'out of sync' and your program hangs.

The following suggestions:
1) Make your interrupt routine faster. Especially every 100th loop is a killer, maybe you can spread this over multiple calls?
2) Add a timeout to your protocol, so it can recover from errors.
3) Remove the setup_uart() and set_uart_speed() calls, you don't need them as #use rs232 is setting everything ok. You only need those functions if you want to change a setting during runtime.

Carlo

[jroth]

Thanks.

I optimized the interrupt loop and have been able to get the program to run at 57600 baud.