Pure two-tone test signal on spectrum analyzer

I needed to get two-tone signal out of a Rigol DSG3060-IQ RF signal generator. Seems like a simple test signal that is needed for most RF testing applications, but boy was I wrong. There was no UI element to get two signals out at the same time, so the solution had to be to use arbitrary waveform generator. This takes waveform, generated in computer and plays it. Works as a basic software defined radio with IQ signal as input.

First I tried to do the official way: downloaded Rigols official Ultra Sigma and Ultra IQ Station. Ultra IQ Station was the most bug filled piece of software that I have ever seen. It has button for multi tone signal, but it gives unrecoverable error each time you think about changing a setting.

.arb file format itself

After ditching this, I moved on to the next logical step - reverse engineering the .arb file format. It seems to be somewhat of a standard, at-least in Rigol equipment, but there is no documentation on the version used in RF signal generators. After bunch of reverse engineering I figured out that it is a binary IQ dump with some headers. File starts with some binary header, then there is JSON metadata that tells the signal generator the sampling rate and number of samples, and then the IQ portion itself has 2 bytes per sample. That is pretty typical for 14 bit DAC based system. But the problem. We discovered that the binary data is somehow protected. Binary header of the file has timestamp or random seed in it and all the data has been changed with that key. Generating .arb file from the same IQ file gives different binary content each time. Unfortunately that means that it is not trivial to just add a header to your otherwise good IQ files. That also means that any .arb file generation has to rely on Rigol provided tools until somebody bothers to reverse engineer the keying.

Short example ARB file:

GÖÌÚARB.Ð...Ò......÷....óãà0.Oö.Ä|.`èÿ.g."è.®r\@}Hÿp.àL.!..3..G£Ñ`.û.oÿÀª.tð.¾(88B!.÷\ÿ.".¹..Òoê8°À¬..dE..98P.......Ù..| WÖ÷........Wæ3.........0..|....L..|isÄ÷L..|.Â.Ú....L..|©KÀ÷.PÖ÷....Þæ3.øè3.snÄ÷....²...{Clock:20000000HZ}{ModulationMode:Digital}{Value:0.874303}{Num:8}{DataSource:PRBS9}{Length:16}{SymbolRate:20000000Hz}{ModulationType:32QAM}{FilterType:Raised cosine}{FilterCoefficients :1.000000}{ImpulseLength:1}{OverSample:1}Z..7%í%LR¯Sâók{..ÿÒ.úÁ.Õ.*¼jp{.@

Official ConvDat2Arb.exe utility

While trying to figure out file formats we found a poorly documented command line utility called ConvDat2Arb.exe from one of the Rigol installers. As the limited help said it takes IQ data in either CSV or binary format and gives out .arb files. No information on data format, only the type. First we figured out the CSV. Contrary to the name "Comma-Separa Values" the tool can't handle any commas... Each value has to be on separate line and no commas can be in the file. The tool takes your maximum and minimum values and scales them properly for the signal generator DAC. It doesn't matter if values are between 0-100 or 0-100 000, it still works. It also works with negative numbers, that are converted to negative IQ voltages. I used it to generate my two-tone test signal and both, the IQ outputs and RF output worked exactly as hoped.

Secondly I tried to figure out the binary format. I knew that output file format had 2+2 bytes for each sample. And I knew that file sizes go down 4x when converting with the converter. After some testing we figured out that the input data has to be in long format. So 8 + 8 bytes per sample. With is a bit unfortunate since GNU Radio doesn't support this data type out of the box and separate converting is needed. With binary input negative numbers also don't seem to work. So the signals have to biased up from 0. But as with CSV the input file is scaled to match the signal generator and this is not something the user has to think about.

Internal memory glitches

Rigol DSG IQ glitch from internal memory

With this my adventures in getting signal I need out seemed to end. But no, there is more. After testing the signal from USB stick and everything working I decided to copy the test signal to internal 1 GB memory. USB sticks tend to get lost and this is a much used test pattern. After running it from internal memory I noticed that running same file from internal memory produces endless amount of glitches in IQ outputs. So using the internal memory should be always avoided.

While putting random data in the .arb files I managed to brick my signal generator for some time. So be careful when doing further research.

IviGui RF power meter measurement

Lab where I work lately bought a Keysight U2001A RF power meter. It was an expensive, but is a nice device. Since the software that you have to use was subscription only I wrote my own GUI to take measurements.

As all USB connected measurement devices, this power meter needs a software to display the measurements. As it turns out, the only supported GUI for this and many other similar devices is Keysight BenchVue. BenchVue is a good and big tool for big labs, but comes with its problems. As many electronics software - it is Windows only. It also is around 1GB big and has its own license servers. Also you have to pay subscription fee. After buying measurement device for multiple thousands of euros, you also have to pay monthly subscription fee to use your legally owned measurement device.

All this seemed a bit much for me, to just take simple RF power measurements so I wrote IviGui. Most connected devices provide a way to program scripted test automation. This particular device uses USBTMC. So I took python-usbtmc library and wrote a simple GUI on top of that to measure and display this measurement continuously.

Keysight U2001A in measurement

I also added some features that I felt, were missing in the gigantic BenchVue application. Like a blinking dot to indicate when an actual measurement is updated. Also a clear and simple way to do relative measurements to subtract cable and attenuator losses.

At the moment the GUI script only works with this device and doesn't offer all the functionality of the device. But I'm sure it can be used with other similar USB RF power meters with only minor changes.

School is over, I'm finally able to do some work again.

I had some PCB orders to make. So, script I have been wanting to make for a long time - KiCad gerber renamer. It is a shell script that takes gerber files that KiCAD produces and renames them to match ITEAD or Seeedstudio file requirements. Really handy. Works with multiple versions of KiCAD (daily and older).

GET IT!

FreeCAD Sketcher view

FreeCAD is an open source freeware 3D CAD program. I needed some 3D models for my CubeSat battery post and it looked user friendly, so I gave it a try. I have previously used Blender3D, Solid Edge, SolidWorks, Google Sketchup and HeeksCAD. So I have some experience with 3D modelling and no experience with formula generated modelling (like many open source CAD programs tend to be).

First impression - where are ALL the buttons. Then I discovered workbench switching - makes sense now. Couple of small tutorials later and I was able to model simpler things.

It has usable 2D sketching tool and possibility to modify 3D models. Also almost normal assembly making ability.

Like all open source tools - the GUI is still a bit rough from the edges. SolidWorks style constraints are a step to the right direction, but they are still a bit buggy. For example clicking on the constraints is pretty challenging. I modelled CubeSat PCB after CubeSat Kit PCB Specification and made some simple assemblies. Pictures are in the last post and here are the models.

So I use MSP430FR because it has FRAM. But it also has 1 kB of SRAM and all the variables and the stack is in the SRAM. That is no fun I want my data do be safe and maybe even survive a power off.

First idea is to use static addresses. But that is super bothersome and error prone to more than 2 variables.

Fortunately there is linker command file in the Code Composer Studio project - lnk_msp430fr5739.cmd. I found lines 119 to 121. They say that variables and stack should be in ram. One simple hack later and:

.bss       : {} > FRAM          /* GLOBAL & STATIC VARS */
.data      : {} > FRAM          /* GLOBAL & STATIC VARS */
.stack     : {} > FRAM (HIGH)   /* SOFTWARE SYSTEM STACK */

Checked it out with debugger, and indeed. All variable memory locations were in FRAM area (device datasheet, pg 23, Table 6. Memory Organization).

Fun - but.. but... there are restrictions that apply!

1. FRAM speed is 8 MHz

So if your MCU is faster then all operations with memory will be delayed a bit. There is some caching, but my application run at 8 MHz and everyone else's even slower so whatever.

2. FRAM is non volatile

You can be extra sure that all your variables contain random values at startup. Or - not random, but the same that they had on last power down.

3. FRAM is not indestructible

Datasheed explicitly warns about it. But also it says that it will survive 10¹⁵ writes. Most read and writeable variable in my code is i. It is the counter in inner loop. It gets read or written (endurance wears with reading too, it has destructive reading) about 5 times per loop. Inner loop runs 100 times and the outer loop runs at 600 Hz. So my code writes the variable at 5*100*600 Hz = 300 kHz. With this much use the FRAM will be destroyed in - 105.62 years... So.. I think I'm good.

I still hate registration walls and US export approvals. I upload so you wouldn't have to see them. This time: Grace – Graphical Peripheral Configuration Tool. Also, my server seems to be faster than ti.com...

http://jaanus.tech-thing.org/img/grace_setup_2.0.0.00017.exe

I hate registering to pages to get information or downloads. I really hate it. This is one of the two reasons Arduino has been so popular - it's open source and you can do whatever you want with it.

So I have a Texas Instrument FRAM development board MSP-EXP430FR5739 that I can buy here, in Estonia with no questions asked. I want to program it - I have two choices: commercial Windows only IAR or cross platform Eclipse plugin Code Composer Studio. Off course I try to download Code Composer Studio. But no, I can't. I have to fill out some stupid U.S. Government export approval. WTF. I can get the board and all your high tech chips with no problems but downloading plugin to my Eclipse requires a export document. You have to be crazy. So, if anyone wants it. I host a mirror now:

LINUX DOWNLOAD

WINDOWS DOWNLOAD (1.3GB)

Edit: Windows installer is now offline one so it would work.

For one part of ESTCube-1 satellite we need to compile big and changing project to our AVR ATmega. Just using AVR Studio doesn't fit because we have people with linux (me), windows and mac in our team. The other common way - using hand modified AVR Studio Makefile doesn't fit either because it is prone to errors and irritating when working with tens of files. Lets use build automation tools! First though was Makefiles. After half a week of studying about the cryptic ways of Makefiles it was clear that it is not the right way. So lets use something simple (simple usually means python) - Scons. It is a tool very similar to Makefiles. But instead of writing cryptic and non debuggable pseudo code we can use python! I worked on it for couple of days, finally taught myself Scons with really really simple example and then copied the structure to real project.

The base for the code comes from some weird german wiki.

#coding=utf8

env = Environment()

mcu = 'atmega1280'

if len(COMMAND_LINE_TARGETS):
Target = COMMAND_LINE_TARGETS[0]
else:
Target = ''

# Optimization level, can be [0, 1, 2, 3, s].
opt = "2"

env['CC'] = 'avr-gcc -mmcu='+mcu+' -O'+opt
env.Append(CCFLAGS = "-Wall")
env.Append(CPPPATH = 'EPS_Library')

# Make elf
env.Program(Target+'.elf', [Glob('Library/*/*.c'), 'Tests/'+Target+'.c'])

# Make hex
env.Command(Target+".hex", Target+".elf", 'avr-objcopy -j .text -j .data -O ihex $SOURCE $TARGET')

# Show memory usage
env.Command(None, Target+".hex", "avr-size $SOURCE")

# Program the thing
env.Command(None, Target+".hex", 'sudo avrdude -c dragon_jtag -P usb -p m1280 -u -U flash:w:$SOURCE')

To run it just write:

scons Test0

scons Test0

Where "Test0" is the name of the target file in the Tests directory.

The code compiles the test and library to elf with avr-gcc. Then the the elf to hex. The avr-size then shows the size. Finally avrdude, using AVR Dragon JTAG erases the device and flashes fresh .hex file. For clean command the target must also be specified.

scons Test0 -c

scons Test0 -c

Qt4 grass physics engine

I got a little bored when visiting my parents and last place I tried to go to work didn't want me because of my poor C++ skills. So best way to practice OOP in C++ and familiarise myself with the syntax is to make a project. My idea of project - a 2d physics engine for grass... It had GUI, OOP and was simple enough for one evening project.

I found very good tutorial from site http://zetcode.com/tutorials/qt4tutorial/ . Surprisingly GUI programming in C++ was exactly the same as in Java or in Python. The C++ syntax was easy too because I had seen it for years, and finally I understood why it was like that all along. Also qmake is very useful, must learn that more.

Secondly I realised (again) that random function is very useful in games, and more or less useless in everything else. And finally - the GUI worked at 100FPS, I love fast reacting GUI. Damn shame that my computer is slow/Intel Atom.

Source/binary: http://jaanus.tech-thing.org/img/c++_qt_test.tar.gz

It is sad how all CADing is still be done with proprietary software. Like: I use SketchUp, it's easy to use, but the full version costs and it's not made for technical drawing. So searched in the web and found many free CAD programs. But all of them were unusable for Sunday mechanic. Most of free linux programs had so steep learning curve or were just plain retarded from modern point of view. The only interesting thing I found was BlenderCAD. This is a project that tries to make plugin that converts existing and good 3d software Blender3d into CAD program. I have used Blender and it is a very good program for drawing, also many people(like Open Source Ecology guys) already use it for CADding. Unfortunately I can't talk more about it because I wasn't able to install it - a classical problem with non-pro linux users.

First model in HeeksCAD

Later, while cruising around in RepRap wiki I found HeeksCAD. There were some problems with installing. I followed http://code.google.com/p/heekscad/wiki/UbuntuInstallation , but used too new Ubuntu for it, so I temporarily broke package system on one computer, couldn't install on other. They really should get it on the Ubuntu repositories. I would have made a donation to them with that comment, but they donate button is not working on my computer.

After installing I found myself from 3d space, much like SketchUp. But differently from SketchUp HeeksCAD is more number based CADing program and less like drawing. I watched couple of tutorials and tried a bit of modelling and I approve. I think this is THE CAD that I have been looking for. It even has CAM built in it. On the downside I couldn't find measuring button and the undo isn't working half of the times. But it still is very promising project. It will be one of the biggest hobby CAD programs around.

The second day started with some talks. First one was about human sensors and how human fuses its sensor readings together. Very fascinating was to listen analysis of human sensors like humans were robots. They discussed how humans find out the error percent of all of their sensors and how they measured all those constants.

The second talk was about robot companions (http://robotcompanions.eu/). Its a long term project to make robots for home use and prepare humans and robots for cooperation. Again, pretty interesting. After the talk it got funny questions, like "Won't robots replace humans?" and the answer was something like "We need to overcome these kind of perception." in short and blunt - "Yes, so what". You got to love robotics.

Then was some research opportunities talks, some interesting stuff but way out of my league/interest. And then - 15 minute oral sessions. So, many good topics and bad performers, typical nerds. Papers I remember form there were "Background Elimination Technique in the Structured Light Systems for Dynamic Environments" - about how to eliminate background from moving projector/camera systems. The example of this would be kinect. Another interesting one was "Robust Localization of Furniture Parts by Integrating Depth and Intensity Data Suitable for Range Sensors with Varying Image Quality" - a paper showing ways to localize furniture, useful for future service robots.

After lectures we went to Centre for Biorobotics. They showed female Fits.me robot, UAV and some robot fish. Most of the tour time we just hang out in electronic lab and talked about microcontrollers..

It ended with Interactive session/reception. Finally something on my level and something innovative. I talked with couple of people there (noticed that my English is awful). One of the most interesting papers was named "Magnetic field-based SLAM method for solving the localization problem in mobile robot floor-cleaning task". Basically it showed that indoor magnetic fields are good enough to do mapping using them. The sensor for this is low cost and can be integrated with cheap odometry and cheap low-range distance sensors, creating localisation good enough for most of indoor robots. Very fascinating, the larger scale mapping device was the thing missing from low end indoor robots.

Also worth mentioning is paper "Localization of Intelligent Ground Vehicles In Outdoor Urban Environment Using Stereovision and GPS Integration.", about integration of optical flow and GPS (if you already have 3d map). Also "Navigation strategy and path planning for autonomous road crossing by outdoor robots" - the first mentioning of smaller mobile robots in urban environment, a field that sounds very interesting to me. Also couple of interesting devices, described in papers "Development of an intelligent object for grasp and manipulation research" - small cylinder with massive amounts of tactile sensors and "Vibrotactile Display for Hand-held Devices and Its Preferable Pattern Generation" - the first step towards touchscreens with touch feedback.

This is very awesome convention, it is funny to read paper names what are longer than ten words. Also the specialization of the papers, its mad. I'm sure I will get more used to all this academic world but for now, I'm still first year student.

Flying between planet and its moon

So about a year ago, I started making HTML based game. Technology seemed ready enough and interesting so I thought to give it a try. After a bit of trying it came out that it is very possible to build real time space shooter game using only JavaScript and canvas element, in your face Adobe Flash. For now it has reached version 0.18. It features fast flying, realistic space physics, moving planets, minimap and more. Just go to page http://space.tech-thing.org and start flying. Goal so far is to have fun exploring the space and trying to control the spacecraft.

For graphics I chose Raphael JS library and for server communication and other tasks - JQuery. Space shooter game was selected because I like making physics engines, not games. The version number is so small because I have enormous ambitions, so about one day spent programming pushes the number about 0.01 forward. At the current version (0.18) there is only 4 planets, but I will fix it when my brother (the artist) draws more planet and sun pictures. More planets, naming, changing view size and more spaceships will be in version 0.2.

All planets, spaceships and their positions are stored in a mysql database. Cron job moves the planets once a minute. So they are moving, just so slowly that you can't see it. Game sends a query for planets every minute and, if you fly forward. For other ships I use long polling. Unfortunately Apache+PHP+MySQL isn't very fast combination. Or - they are fast but not real time. So that's why current FPS for other ships is - 5. I'm planning to move to faster server, don't know yet to what. First thought is Google appspot and Python because.. I'm familiar with it.

So yea, my little game project. I hope you have fun.