Hi everybody, Let me report current progress of this thread. Now my matlab program can detect and correct sampling and carrier (both phase and frequency) error for dongles, and show the sampling phase difference between two dongles (at 8X oversampling rate) after correction. You can try it by yourself here : https://github.com/JiaoXianjun/multi-rtl-sdr-calibration Some runs results have also been attached here. (time location of FCCH SCH and BCCH have also been shown there in different color) In the attached pictures, you may find that from the GSM frame point of view those FCCH, SCH and BCCH from two dongles seem almost in the same time location. (this needs to be verified further by check/demodulate information carried in SCH and BCCH, which is still under development). But if we inspect the time location in a 8X oversampling scale (zoom-in), there is a fixed sampling phase difference between two dongles (after correction respectively). Unfortunately, this fixed sampling phase difference may vary from run to run (one rum means we restart rtl_tcp). Fortunately, it remains fixed in a single run. This offers us a opportunity to track continuous input I&Q (don't re-run rtl_tcp!) from multiple dongles, and needn't worry about phase/sampling jumping/discontinuous things. Why do carrier and sampling error estimation and correction respectively? Inspired by James Peroulas james@peroulas.com (author of Lte-Cell-Scanner), for example, when there is a non-coherent up/down-converter used outside dongle (like MMDS LNB used by Omri Iluz omri@iluz.net to detect 2.4GHz band signal), there won't be strict relationship between sampling error (which is in baseband) and carrier error (which may be located in inside dongle or outside dongle mixer). But for my in-hands dongles, sampling error and carrier seems have relationship. This may be because they have common clock source inside dongle. Which I detected by my matlab program gsm_sync_demod.m: dongle 1: sampling PPM 27.1739, carrier PPM -28.513 dongle 2: sampling PPM 116.3043, carrier PPM -118.4247 In my program a positive sampling PPM means sampling clock is slower than standard/ideal source (in our case, this source is GSM base-station). If LO of mixer uses the same clock as sampling, that may cause a lower LO in mixer/down-converter. I guess the LO frequency of mixer in dongle is higher than input antenna RF frequency, which means for mixer: output = LO - input. Thus a lower LO generates a lower input into baseband, that's why I get a negative carrier PPM (negative carrier PPM means detected carrier frequency is lower than standard/ideal source). I don't know if my guess is correct. Anyone is familiar with the hardware architecture of the 820t dongle? BR Jiao Xianjun On Wed, Jan 8, 2014 at 5:23 PM, jdow <jdow@earthlink.net> wrote:

The last I looked rtl_tcp did not allow full control over the dongle. But, then, to do what I really want has required bypassing rtlsdr to go directly to the usb library so I can access the underlying USB system directly for a specific feature. (A feed through feature in rtlsdr's shared library would be nice when trying to nail down the dongle's identity for restoring context on program startup.)

(Building an SDR program without multi-threading is rather like trying to build a bicycle without chains or cables. It can be done. But it's a pathetic tool when you get done with it. Erm, but I do note gear driven bikes need less maintenance under adverse conditions. Been there done that - MANY decades ago.)

{^_^} Joanne/W6MKU

On 2014/01/07 23:32, Jiao Xianjun wrote:

...

Good idea.

Thanks for your advice!

But that seems we will do some repeating works.

rtl_tcp already use mutiple buffers and multi-threading. So why not use it directly as a reliable relay.

On Tue, Jan 7, 2014 at 1:51 AM, jdow <jdow@earthlink.net <mailto:jdow@earthlink.net>> wrote:

Have you tried using ping-pong buffers? Process on one buffer and receive to another. You may have to use multi-threading to make this work nicely.

{^_^} Joanne/W6MKU

On 2014/01/06 01:34, Jiao Xianjun wrote:

Hi,

Today I did some experiments using CW signal which is generated by signal generator. The conclusion is a little bit sad.

sync read and UDP lose samples/data high probably: 1. If there are some other operations (which costs some time) between two successive sync reads, some samples will be lost. 2. Some times, UDP packets are just lost.

So, seems that I have two choices: 1. struggle to use async read mode. 2. use rtl_tcp utility directly, which is offered with rtl-sdr code. This program use async mode and TCP, which has avoided the two shortcomings I addressed.

I will try the 2nd method, and try to move on with calibration.

BR

Jiao Xianjun

On Sat, Jan 4, 2014 at 8:34 PM, Jiao Xianjun <putaoshu@gmail.com <mailto:putaoshu@gmail.com> <mailto:putaoshu@gmail.com <mailto:putaoshu@gmail.com>>> wrote:

Hi,

I have questions on usage of rtlsdr_read_sync(dev, buffer, out_block_size, &n_read):

For example, if sampling rate is 1Msps, and out_block_size is 1000000, question is:

1. the rtlsdr_read_sync() will cost 1s exactly? Or is there any lower level device/driver buffer, which perhaps feed rtlsdr_read_sync() with history data quickly and makes rtlsdr_read_sync() return in a time shorter than 1s?

2. in this infinite processing loop: while(1) { rtlsdr_read_sync(dev, buffer, out_block_size, &n_read); processing_function(buffer); // let's assume that this cost 0.001s } Those samples during the 0.001s processing period will be losted or not? Is there any method to not lost them?

Thanks very much!

BR

Jiao Xianjun

On Thu, Jan 2, 2014 at 9:25 PM, Jiao Xianjun < putaoshu@gmail.com <mailto:putaoshu@gmail.com> <mailto:putaoshu@gmail.com <mailto:putaoshu@gmail.com>>> wrote:

I see some UDP packet performance issues in my laptop (but not in my desktop computer). Will the common (interleave multiple receives) UDP link helps?

The issue is that fread for the second dongle in matlab get less data than expectation sometimes. Seem that fread for the first dongle works well.

------------------------------__---------------------------- --__--------------------

From: Sdr Guru <mailto:sdrguru1@gmail.com <mailto:sdrguru1@gmail.com>> Sent: 2014/1/2 5:39

To: osmocom-sdr@lists.osmocom.org <mailto:osmocom-sdr@lists.osmocom.org> <mailto:osmocom-sdr@lists.__osmocom.org

<mailto:osmocom-sdr@lists.osmocom.org>>

Subject: Re: How to get IQ samples from multiple rtl-sdr dongles in asynchronized manner?

Hi

rtl-sdr-relay Some of the recommendations. Please add PPM error calculation, exactly like new rtl_test -p but multiple receivers simultaneously. It provides immediate information if something is wrong with USB or dongles. https://github.com/keenerd/__rtl-sdr/commit/__ b5f89dcf40463130e717b6c9bb3a39__a3c8b9535f <https://github.com/keenerd/rtl-sdr/commit/ b5f89dcf40463130e717b6c9bb3a39a3c8b9535f> https://github.com/keenerd/__rtl-sdr/blob/master/src/rtl___test.c

<https://github.com/keenerd/rtl-sdr/blob/master/src/rtl_test.c>

Please add automatic eeprom PPM calibration https://github.com/keenerd/__rtl-sdr/commit/__ ecf267737ca52f5005b7a12a352307__e8cd763ed6

<https://github.com/keenerd/rtl-sdr/commit/ ecf267737ca52f5005b7a12a352307e8cd763ed6>

default sample rate 2.4M (28.8/12) or 1.2M (28.8/24), probably lower jitter MAX_NUM_DEV 4->16 :)

Some nice to have features. ip binding multicast support one common (interleaved) stream of all the receivers timestamped stream

I'm trying to convert MATLAB script to Ocatve.

SG

On Mon, Dec 30, 2013 at 9:38 AM, Jiao Xianjun <putaoshu@gmail.com <mailto:putaoshu@gmail.com> <mailto:putaoshu@gmail.com <mailto:putaoshu@gmail.com>>> wrote:

Hi guys,

For the multiple dongles synchronization in signal level instead of bits/packets level, I setup a working repo in github, and write a initial demo framework. See below:

https://github.com/__JiaoXianjun/multi-rtl-sdr-udp-__relay.git

<https://github.com/JiaoXianjun/multi-rtl-sdr-udp-relay.git>

You may find information and instruction of demo quickly by reading the README.

My initial purpose is performing in-fly calibration for multiple dongles according to some pre-known signal (GSM, ADS-B?) to let them work together coherently.

An ideal scheme may be that we should generate a very narrow band and very week signal in (or just located at the edge of) target working band of dongles, and perform the software in-fly calibration in background (or driver level). This would be user friendly.

I know it is far from final state currently, and many things are not clear yet (See TODO). But please join me if you also think this is a good idea. Just check out the demo and run it to have a look.

Currently I just test the demo in Ubuntu-Linux.

BR

Jiao Xianjun