Problem communicating via UART in non-blocking mode with higher baud rates

[Akhil Bhargav Josyabhatla]

Hi,

I am trying to communicate with a Wi-Fi module external to nRF52840 via UART. I am using the nrf_serial_read() API. I have tried different values for UART baud rate from 115200 to 921600 but nothing works in the non-blocking mode(timeout_ms = 0).

The only working combinations between 115200 and 921600 are:
Baud rate                   Timeout         

115200                      125ms
230400                      100ms

Q1) Is the timeout necessary to make the UART communication work with a high baud rate?

Q2) I need to achieve a baud rate of 921600 for a smoother application with a minimum timeout in ms. Is there a way to achieve it? 

Thanks,
Akhil

[MartinBL]

Hi,

 

1. How does it not work? 
2. Do you check the return code of nrf_serial_read()?
3. Does it return an error? 
4. Do you have a logic analyzer?

I don't think the timeout should matter. Note though, that you won't achieve 921600 exactly, but 941176,a so you are likely to get framing errors which can be problematic. 

 

 

[Akhil Bhargav Josyabhatla]

Hi MartinBL, 

I got the nrf_serial_read working. I see that the communication is successful but I get framing errors and cannot receive complete data even with 115200 baud rate at timeout = 0.
Same is the case with all other baud rates till 921600. 

But, for baud rate 115200, I see reliable communication without any framing error for timeout = 125ms. 

I guess you're right, the timeout shouldn't matter but the timeout is playing some role here and I am confused about that.

[MartinBL]

Hi,

So the Zentri eval board works with all baudrates, but if you connect a module directly to the nRF52 only a few baud rates work?

It could be that the nRF52 isn't able to provide an accurate enough baud rate for your module. As you can see in the specification, the baud rate you want isn't always the baud rate you get. 

Desired baud rate	Actual baud rate	Error
9600	9598	-0.02 %
14400	14414	0.10 %
19200	19208	0.04 %
28800	28829	0.10 %
38400	38462	0.16 %
57600	57762	0.28 %
76800	76923	0.16 %
115200	115942	0.64 %
230400	231884	0.64 %
250000	250000	0.00 %
460800	470588	2.08 %
921600	941176	2.08 %

In addition to the errors here, you will have to factor in the accuracy of the HF clock in your nRF52. This is because the digital logic in the UART is driven by the HF clock and the clock accuracy will hence directly affect the baud rate accuracy. Unless you tell it otherwise, the HF clock is generated by the internal RC oscillator (HFINT) with an accuracy that could be as bad as +-6 %. You can try to force the HF clock to utilize the external crystal (HFXO) as a source instead to increase the accuracy (typ 40 ppm or less). You can turn on the crystal like this:

NRF_CLOCK->TASKS_HFCLKSTART = 1;
while(NRF_CLOCK->EVENTS_HFCLKSTARTED == 0)
{
    ;
}

or with the function sd_clock_hfclk_request() if you are using a Softdevice and BLE. 

I suspect that your Zentri Eval board has an FTDI (UART-to-USB) chip on board with better baud rate accuracy, and maybe that is why they seem to work. 

[Akhil Bhargav Josyabhatla]

Martin,
Thank you so much for the info, that was really helpful. I have a few more questions regarding calibrating the Internal RC.

My hardware setup has 2 nRF52840s- 1 Main and 1 Secondary connected via SPI. The Main chip alone has the external crystals connected and Secondary Nordic (Without external crystal) connected to the Zentri module over UART. Here's a block diagram:


Do you think there's a way for me to calibrate the Internal RC of secondary chip with main nordic chip's clock? Doing this, I believe will help me to make the communication much more reliable at a baud rate= 921600. 


Best,
Akhil
 

[MartinBL]

Happy to help.

Do you think there's a way for me to calibrate the Internal RC of secondary chip with main nordic chip's clock?
Unfortunately this is not possible. And note that if you plan to use the radio on your Secondary nRF52840 it will need a HF crystal too. Radio protocols like BLE for example, needs very accurate clock sources (<40 ppm). 

[Cecylia]

MartinBL in the specification that you gave, it shows that 1M Baud has 0 error.  But, if we don't use an external HF crystal, is it possible to achieve that? Or will that still be affected by the +-6% error?

We tried with 1M and it seemed to be more reliable for a longer time, but eventually it breaks down and we seemed to have missed some bytes.  Is it very likely that it's caused by this error?

[MartinBL]

Or will that still be affected by the +-6% error?
Yes. The UART baud rate is derived from the HF clock so your baud rate will have the same inaccuracy that your clock source have plus the error shown in the table which is a result of how the UART scales down the clock signal.