Frequency Accuracy

Discussions related to schematic capture, PCB layout, signal integrity, and RF development
Post Reply
uhakansson
Posts: 3
Joined: Wed Jul 15, 2015 9:17 am

Frequency Accuracy

Post by uhakansson »

Hi everybody,

I am new to the bladeRF. I have been working on Ettus B210 Radios for awhile and wanted to try out the bladeRF, so forgive me if I am asking the obvious.

I am trying to figure out what the frequency accuracy of the bladeRF is. I want this piece of information so I can can calculate what is the maximum expected Carrier Frequency Offset (CFO)?

I am confused by three different pieces of information given on the bladeRF product

First the product description at https://www.nuand.com/blog/product/bladerf-x115/.

Highlights:
40ppb (0.040ppm) factory calibrated VCTCXO

Technical Specifications:
16-bit DAC factory calibrated 38.4MHz +/-1ppm VCTCXO

Then I looked at the kalibrate-bladeRF instructions at https://github.com/Nuand/kalibrate-bladeRF
"The bladeRF has an on-board VCTCXO (Voltage Controlled Temperature Compensated Crystal Oscillator) that is factory calibrated to within 20ppb. This calibration is partly found by reading the VCTCXO's output frequency with a frequency counter as the on-board trim DAC's output voltage is modified programmatically. The goal of the calibration procedure is to get the VCTCXO's output frequency to be as close to 38.4MHz as possible..."

What accuracy (ppb or ppm) number should I use to calculate the maximum expected Carrier Frequency Offset (CFO) at a particular frequency?

1 ppm? Then the CFO at 900 MHz is 900 Hz.
40 ppb? Then the CFO at 900 MHz is 36 Hz.
20 ppb? Then the CFO at 900 MHz is 18 Hz.

Can anyone explain/clarify this?

Regards
Urban Hakansson
bpadalino
Posts: 303
Joined: Mon Mar 04, 2013 4:53 pm

Re: Frequency Accuracy

Post by bpadalino »

The VCTCXO is 1ppm so you should get that stability over the entire temperature range.

At the factory we tune to be within 1Hz at 38.4MHz at room temperature.

From a fresh board, you should be well within 0.5ppm and probably more like 0.1ppm - depending if you have your board sitting in the sun or not.

Hope this helps. Let me know if you need more clarification.

Brian
uhakansson
Posts: 3
Joined: Wed Jul 15, 2015 9:17 am

Re: Frequency Accuracy

Post by uhakansson »

Brian,

Thanks for the reply, but I am still a bit confused.

Can you help me understand how to interpret the claim made in the product specification the bladeRF having a "40ppb (0.040ppm) factory calibrated VCTCXO". How does this relate to what you just stated?

I am basically trying to figure out if the bladeRF has a VCTCXO accurate enough to fulfill the GSM BS requirements which is 50 ppb, so I could run for example openBTS within specs, see for example http://openbts.org/w/index.php/Clocks.

I found this interesting article on the internet, "SDR showdown, hackRF vs. bladeRF vs. USRP" by a Taylor Killian/
http://www.taylorkillian.com/2013/08/sd ... -usrp.html
The bladeRF Kickstarter campaign was successfully funded and all of those units have been shipped. The bladeRF provides access to a large portion of the radio spectrum along with a large FPGA and fast USB3 link. The developers have invested heavily into their clocking architecture and provide a VCTCXO that has been calibrated within 50 ppb. All of the components were designed to be in sync so that no ClockTamer is necessary.

This article was actually what made me curious about the bladeRF in the first place, an SDR with a on-board VCTXCO, not needing an external clockTamer, or board-monted GSPDO, calibrated within 50 ppb.

In your opinion, is this article incorrect with regard to what it says about the clocking architecture and the VCTCXO?

If the Frequency Accuracy is only 1 ppm, then the bladeRF does not meet the 50 ppb GSM requirement. As a comparison The USRP B210 has an on-board VCTCXO that has a 2 ppm frequency accuracy but it does accept a board-mounted GPDSO that can bring the frequency accuracy to 75 ppb without GPS lock and to 1 ppb with GPS lock, so it could run openBTS wuithin spec. From what I understand the bladeRF does not even have the ports to accept 10 MHz and 1PPS external references from for example clockTamer or a GPSDO. So the on-board VCTCXO has to suffice, and it would, if its accuracy was 40 ppb, but not if its accuracy is only 1 ppm.

I apologize if was rambling on here a bit, but I wanted to be as clear as possible as to why am I still a bit confused.

Again, thanks for you taking the time and read my questions/raised concerns.

Regards
Urban
bpadalino
Posts: 303
Joined: Mon Mar 04, 2013 4:53 pm

Re: Frequency Accuracy

Post by bpadalino »

The 40ppb comes from the calibration at room temperature during factory calibration. If, lets say, you were off by 1.5Hz at 38.4MHz during that calibration, then your total offset would be 1.5 / 38.4 = 0.039ppm. As the temperature goes up, air blows on the VCTCXO, or any other number of factors, the crystal will deviate up to 1ppm. If you keep yourself at the same temperature, then the crystal will be very stable.

We don't have the ability to lock to a reference right now, that's correct. There is a little header going into the FPGA which can accept a 1pps and 10MHz clock, so the hardware supports the ability to tame the oscillator but the code to set the trimdac and do the long term averaging of the 1pps/10MHz signals over time versus the 38.4MHz oscillator is not written.

If you want guaranteed stability to run a GSM BTS over temperature for years and never deviate outside of 50ppb, then the current oscillator is not suitable. If you want to setup a small BTS for local testing on a lab bench, then the radio is very much well suited for this operation.

I hope that cleared up any confusion you may have had. Please let me know if you want more clarification.

Brian
uhakansson
Posts: 3
Joined: Wed Jul 15, 2015 9:17 am

Re: Frequency Accuracy

Post by uhakansson »

I wrote this while you replied so please take this for what it is.

I had an idea where the 40 ppb in the data sheet comes from. 40 ppb is simply the accuracy with wich the trim DAC can adjust output frequency of teh ASVTX-12-H10 VCTCXO, by increasing/decreasing the input to the DAC with +/1.

The output frequency of the ASVTX-12 VCTCXO-H10 has an accuracy of +/- 1ppm, which leads to the LMS_CLK(controls the LO on the LMS6002D), and LMS_TX_CLK(controls the ADC on the LMS6002D) and LMS_RX_CLK(controls the DAC on the LMS6002D) has an accuracy of +/1 ppm. Carrier Frequency Offset (CFO) and Sampling Frequency Offset (SFO) would be calculated using +/- 1ppm.

This is how I calculated it.

16 DAC: 0...65535 input (65536 range), 0.004 V - 2.494 V output (2.49 Volt range).

ASVTX-12 VCTCXO-H10 input control voltage 0.4 - 2.4 V (2 Volt range)

(2 Volts / 2.49 Volts) * 65536 = 52639 DAC input range that actually has an effect on the ASVTX-12 VCTCXO-H10 output frequency.

ASVTX-12 VCTCXO-H10 Max Frequency Tuning Range: +/-9.5 ppm * 38.4 MHz = +/- 364.8 Hz (729.6 Hz range) .

729.6 Hz / 52639 = 0.01386 Hz / count.

Accuracy/count 0.01386 Hz / 38.4 MHz = 36.1e-9 ~ 40 ppb.

Is this even close to being correct?

If this is true this means the bladeRF does not meet the GSM requirements specifying a +/-50 ppb accuracy for a BTS. Is this correct?
bpadalino
Posts: 303
Joined: Mon Mar 04, 2013 4:53 pm

Re: Frequency Accuracy

Post by bpadalino »

I believe you are off by a few orders of magnitude in your final calculation, but the rest of it looks good.

Being off by 1Hz out of 38.4MHz is 0.026ppm or 26ppb. Being off by 0.01386Hz out of 38.4MHz is being off by 0.000361ppm or 0.361ppb. That's a very, very tight oscillator.

The VCTCXO is not stable enough, off the shelf, to support those requirements - you are correct.

The SMB port on the top of the board accepts a 38.4MHz clock to be in MIMO expansion mode - basically sharing the baseband sample clock from a different board. If you want to be guaranteed to have a 100% stable GSM BTS clock, then I would recommend driving the SMB with the stable 38.4MHz clock of your choice and set your board to "slave" mode. This will drive all the circuitry from your clock source instead of using the VCTCXO on board. The only real requirement is that it is 38.4MHz. I'd use a nice trimmable OCXO. I believe you can get these, in small quantities, for around $150 at the specified frequency and accuracy, possibly even more accurate.

With that being said, if your environment is stable enough (temperature, airflow, etc), then the crystal won't drift outside of the 50ppb range once it is tuned. We have verified this with an E4406A looking at GSM bursts and tracking frequency offset. Once the temperature is stable, and the trimdac is set correctly, we typically don't see the frequency change that much unless the board is sitting in the sun, you put your finger on the oscillator, or you vibrate the board.

I hope this is helpful. Let me know if you have any other questions.

Brian
Post Reply