TMS570 HTU to main CPU memory transfer in dual buffer mode

Chuck,

We have received your question, and one of our expert will be back to you.

Bonjour Jean-Marc,

Thank you very much!

Chuck.

Thank you Zhaohong,

Yes I am going to implement a timeout mechanism using the same signal inputting to another NHET pin. The NHET processing of this signal will obviously use a different register than the one used for the timestamp and signal edge detection.

Since I'm really new to the wonderful NHET world, can you suggest the instructions that I should use for the timeout detection?

Another question: If all buffers A+B are cleared (or set to all 0xF) after reading by the CPU, can I discard the part of checking the OLD Current Buffer Count that you've suggested (check the buffer values instead)? Just want to confirm my understanding.

Thanks again.

Zhaohong,

I figured out that maybe I should use the DJZ instruction coupled with the MOV32 instruction. Is the following sounds?

L00         CNT    { next=L01, reg=A, comp=EQ, irq=OFF, max=0x1FFFFFF }

L01         WCAP   { next=L02, reg=A, pin=CC0, irq=OFF, event=BOTH, reqnum=0, hr_lr=HIGH, data=0, hr_data=0 }

The logic should continue to:

L02     If WCAP detects an edge: next=L03; otherwise: next=L04

L03     Reprogram the register used by the DJZ by using the MOVE32 instruction: next=L04

L04     Execute the DJZ instruction (interrupt will be generated if register reaches zero): next=L00

My problem is I don't know how to implement L02, L03 and L04 using NHET instructions.

Any help would be highly appreciated.

Chuck,

Would you please explain what functionality you want to achieve with the NHET code you posted?

Thanks and regards,

Zhaohong

Sure Zhaohong,

We're building a software decoder for an 1 Mpbs Manchester data stream (clock and data are mixed into the same analog signal). The signal of the stream is variable with different duty cycle in the waveform. The shortest transition changes are every 500 ns (0.5 us) and the longest are 1 us (averaging ~1.5 million transitions/second). To decode this signal with software, one of the solutions is to capture a timestamp and the direction of the transition (failling or rising edge) then analysis it. We expect two words of some 1600 bits each every 10 ms.

It is almost unthinkable to get the CPU interrupted every 500 ns, since the CPU would have other tasks to perform. This is why I want to use the HTU for the transfer between the NHET and the CPU.

We will clock the CPU at 100 MHz, leaving the NHET's peripheral bus VBUS2 running at 50 MHz, and I want to configure the NHET HR=1 (only 20 ns) per HR and the LR will be configured in function of the number of instructions required to do what has to be done.

Implementation wise, I plan to have the same Manchester signal fed into two different NHET pins, one for the actual signal decoder and the other (with the same signal), to the timeout reset mechanism that will be implemented within the NHET. I think that I can handle the algorithm side of the design, but since I'm really new to the TMS570, I lack the knowledge on the implementation using the NHET, hence my support request to your knowledgeable guys (I hope not to spend 2 months to understand all the NHET).

Here is what I want to translated into NHET code:

L00: Start a free-running timer for timestamp capturing purpose (REG_A), goto L01

L01: Search for an edge in pin #1. When found, trigger the HTU to transfer the contents of CF and DF into one of the double buffers in RAM (the HTU will trigger interrupt to CPU once buffer full), goto L02

L02: Start a second timer with predefine counter value for the timeout mechanism (REG_B), goto L03 (obviously, I can't restart this on every loop cycle, how to do that)

L03: Search for an edge in pin #2. When found, reset timeout register REG_B, goto L00

L04: Execute DZJ, when =0, request CPU interrupt, reset timeout register REG_B, goto L00

Your proposed solution is perfectly valid, I believe, but because of the tight timing, I really want to use the NHET to detect timeout.

Can you help?

Hi Zhaohong,

I was wondering whether I should expect further posts from you to help me out ...

Thanks.

Chuck,

You can try the following in your application.

L00         CNT    { next=L01, reg=A, comp=EQ, irq=OFF, max=0x1FFFFFF }
L01         WCAP   { next=L02, reg=A, pin=CC0, irq=OFF, event=BOTH, reqnum=0, hr_lr=HIGH, data=0, hr_data=0 }
L02         CNT    { next=L03, reg=A, irq=ON, max=0x1FF }
L03         WCAP   { next=L01, reg=A, cond_addr=L04, pin=CC1, event=BOTH, data=0}
L04         MOV32  { next=L01, remote=L02, reg=NONE, type=IMTOREG&REM, data=1}

Instructions L01 and L02 are for the capture of the input signal from pin0. L02, L03, L04 is for checking the timeout condition. An interrupt is generated when the count is reset to zero after reaching the max value. Any rising or falling edge on pin 1 (assume that pin1 is connected to pin 0) will set the count value in L02 to 1. You can adjust the max value in L02 for the time suitable for your application. Since NHET is constantly running, you may see the interrupt from L02 for multiple times if the input is inactive for some long time.

When using NHET to capture a pulse, please make sure that the minimum period of the pulse has to be larger than 2xLRP time. The HR field only adds more accuracy of reading. in other words, the rise-to-rise and fall-to-fall time has to be larger than 2xLRP. On the current devices, WCAP sometimes cannot capture HR field correctly for some timing condition. This issue will be fixed in new revision of silicon.

I hope that you have all the information you need for this question.

Thanks and regards,

Zhaohong

Sorry Zhaohong,

Just one point: in L03 data=0x1F, I meant to reset both REG_B and data field of L01 CNT instruction to 0x1F (reset the timeout counter when edge is found). Is there a problem in my understanding?

To reduce interrupts generated by L01 after timeout (on every LR after frist timeout), can I reprogram L01 to irq=OFF, then re-turning irq=ON once an edge is detected again? (I will write more lines of code).

Thanks again.

Chuck.

I got it now. Thank you very much.

Chuck.