Download Radioscape Driver

The Freq-Mite from Small Wonder Labs is ideal for incorporation within an HW-7 or HW-8. (See notes here about installing a Freq-Mite in the HW-8.). In HW-8 Handbook, First Edition (1994), Michael Bryce recommends replacing the MPF-105 front-end in the HW-8 with a 2N4416 for a significant increase in sensitivity. DHRUBOJYOTI DATTA CHAUDHURI Mobile: +141, +916 E-Mail: dhrubdc@yahoo.com, USA B1/B2 VISA Seeking Architect/Sr. Architect level assignments. Download PDF Info Publication number US6A1. US6A1 US10/840,109 US84010904A USA1 US 6 A1 US6 A1 US 6A1 US.

For PC users, Total TV Player is available to download for Windows 8 or later. (2.5 GHz Intel/AMD Core Duo (or similar), 4GB RAM) eyeTV T2 ab €129,95: eyeTV T2 ab €112,95: eyeTV T2 (Guide) £99.99: eyeTV Hybrid (Guide) £119.95. Download this file. 24254 lines (24253 with data), 1.0 MB.

1DABUSB Wavefinder Information - Version 0.5
- 1.2Hardware
- 1.3General software tools
  - 1.3.1Psion's Wavefinder software
- 1.4Custom software tools
- 1.6USB Protocol
- 1.7Other links

Wavefinder is a software-defined DAB radio designed jointly between Psion, RadioScape, Roke, Lost Wax and Real Networks; and marketed by Psion Infomedia circa year 2000. Wavefinder was originally a Windows-only product - the eventual aim of this project is to develop Free/Open Source GNU/Linux software, drivers and firmware suitable for controlling the Wavefinder from an embedded Linux platform such as the BeagleBoard.

In the future it might even be possible to replace the Wavefinder hardware with Open Hardware that takes advantage of modern highly integrated ICs (such as the TEMIC U2731B / Atmel ATR2731 or the Hitachi HD155080 tuner ICs) and using the DSP onboard the OMAP IC rather than 2x5402s.

Todo

Remove any rubbish that's been superseded (ie. stuff we've figured out since).

Hardware

Main section (under shielding on USB cable side)

2x TMS320VC5402 - Texas Instruments 16 bit 5402 DSPs, DVC 5402GGU CD-0AA19DW
ScanLogic SL11R-100 - USB controller / 16 bit RISC processor (User Manual)
2x K6R1016V1C - 64K x 16 Bit High-Speed CMOS Static RAM(3.3V Operating)
ADC1173 EM06AB - 8 bit 3V 15MSPS 33mW A/D converter
MAX5541 - 16 bit serial DAC
LMC6462BIM - Dual Rail to Rail Op-Amp
TL431 - Programmable Precision Reference
2x LCX74PDXP - Dual D-type Flipflop
HCT244 - 8 bit buffer
XILINX XCR3128A - Coolrunner PLD (programmable logic device)
IDT QS3VH245Q XQ0041D - 8-bit bus switch (link)
NC7S04 - Single CMOS inverter (marked 7S04)
48.000 MHz Crystal Oscillator / sytem clock
16.384 MHz device, marked 0047SM1. Reference clock?

The two TI 5402 DSPs provide about 200 MIPS of grunt. According to press releases from TI (link, link), these perform a limited amount of demodulation/decoding of the raw RF signal. More recent designs as used in the ModularTech PCI card, perform much more of the decoding on-board using a TMS320DRE200 (or TMS320C5416) DSP, and less on the host PC.

Radio section (large shield on reverse side)

LMX2331 - Freq synthesiser or PLL (2 GHz & 510 MHz)
LMX1511 - Freq synthesiser or PLL (1.1 GHz)
Hitachi HWSD231 - 38.912 MHz SAW filter (product page, product page)

Other bit (small shield on reverse side)

24LC16B - 2Kx8 Serial EEPROM (prob interfaced to USB controller)
MAX1692 - Low-Noise, 5.5V Input, PWM Step-Down Regulator (marked 1692 EUB)
Si9424DY - P-channel 20V (D-S) mosfet (marked 9424 JZA Y02C)

Power Regulation and Management (Revised 22/11/02)

There is a plethora of power supply devices. It looks like a MAX1692 PWM regulator (rated 600mA) provides a 3.3V supply from the USB 5V supply. This almost certainly provides the power for the SL11R and probably the DSPs and Xilinx chip as well. It also feeds an adjustable linear regulator set to 1.8V (LM1117MPX-ADJ - SOT223 package marked 'NO3A') which probably forms the core supply for the TMS320C5402's. There is also a mosfet (the Si9424) which appears to switch the unregulated power from the mains unit to a ZSR800 8V linear regulator which supplies a ZSR500 5V linear regulator. This regulator chain looks like it supplies the RF section (and, of course, the (in)famous LEDs). There is also a TPS76338 3.3V linear regulator (the SOT23-5 package marked 'PBEI'). This is fed directly from the Si9424 mosfet and here we see a (the?) reason for the Wavefinder's poor reliability. The TPS76338 has a maximum input voltage of 10V according to the data sheet and that is what the loaded unregulated supply provides - this means that this regulator is operating at the edge of its specifications and is liable to fail - or at least it had on mine! The reason whoever designed it used a 10V PSU is that the ZSR800 has a minimum input voltage requirement of 10V. Why the TPS76338 isn't fed from either the ZSR800 or the ZSR500 I don't know - they're both rated at 200mA.

SL11R microcontroller

The SL11R's memory map:

It seems that the memory mapped registers (0xC000-0xC0FF) are mapped directly from USB commands such as 'e6 c0 00 00', etc. The processor is 16-bit, and data words are sent LSB first. This holds true for commands with the first word being in the range (0xC000-0xC0FF). It *may* be that some of the other commands are writing direct to memory.

General software tools

Psion's Wavefinder software

The basis for reverse-engineering the Wavefinder, the Psion software provides the libraries used by any of the Wavefinder control software (e.g. native Psion software, DABBar, etc.). It also provides an API for use by custom programs such as Dabble (extrememly useful when combined with hooking the DLL functions).

Files

rsDSP[ab].bin - contain the DSP code. The data from both of these files is sent to the Wavefinder at startup.
rsCPU*.dll - different version of CPU-intensive code for different processors
ViadabReceiver.dll - lots of stuff, including LED control code

Snoopy/SnoopyPro

This is a USB sniffer, logging all tranfers over the USB bus related to particular devices.

Custom software tools

Dabble

Dabble uses the Radioscape API to control the Wavefinder while hooking some DLLs. This allows the behaviour of the supplied Wavefinder software and what it sends over the USB bus to be studied.(maybe something on interpreting dabble output)

PAD extraction

Kristoff Bonne has written a PAD extractor. pad_decode and tdc_decode. (url) (add more details)

wlights

Controls the Wavefinder's LEDs independently of the Radioscape DLLs.

wboot

Further development of the wlights code so that now it will bootup the wavefinder, download the DSP code, tune it to 'something', and collect all the data returned. No real attempt so far to understand the content of the data streams, other than the semi-obvious.

find_wavefinder() - Now uses some setupapi calls to find the wavefinder driver, based on its GUID. My PC seems to like both {96cb3fae-594e-11d3-b317-00e02914a689} and {a5dcbf10-6530-11d2-901f- 00c04fb951ed} - quite where these come from, I'm not sure, but the {a5dcbf10-... one matches one posted to the list, so it must be wavefinder specific. Hopefully, this code should now find any wavefinder driver, on any USB port ...init_wavefinder() - Trundles through the API calls captured by dabble. (see startup, below)

A few threads are kicked off - one to cycle the LEDS, one to control the frontend, and one to collect the main data:lights_thread() - updates all 3 LEDS every 100ms - takes a 12 bit value, 0x3ff = dim, 0x001=brighttune_thread() - This is where things get serious... A few commands are issued, then 0x800 bytes are returned from wavefinder. This data appears to be the raw output from the ADC in the tuner frontend - plot it, and you'll see. The VIADAB code, then performs a series of DSP functions - applies the Hamming window; fft's etc. I guess this is the tuner synchronisation, and feedback from these calculations gets fed back to keep the tuner locked. It also seems to calulate its BER from this data. This is what we need to understand more to control the fine tuning ourselves.data_thread() - This is also quite serious ... This thread is kept in sync with the tuner thread, and returns blocks of data (0x200 bytes), with a big discontinuity at 0x180 bytes. My guess is that this is the output of the IFFT performed by the DSP chip. (more detail based on wfic) The whole data stream is not returned - so the parameters sent each time round the loop by the tune_thread, must select which what services are being received.Since no sensible data is being sent back to wavefinder in wboot, the data_thread is likely to return garbage. But, if you run dabble, you get the same 2 distinct data streams.

wfic

Extract FIC data from Dabble4 logs, for later processing with FICDEC.

FICDEC

Quick and dirty DSP dissasembler

A partial DSP dissasembler for the TI 5402 DSP code has been written, but further work on this is not expected to be fruitful.

Specifications

ETS 300 401 (AKA 'radio broadcasting systems; digital audio broadcasting (DAB) to mobile, portable and fixed receivers').

(add links to above)

USB Protocol

Startup

PWM Control Register = 0x800f = b1000 0000 0000 1111 which basically means it's set to begin operation, at 48MHz, continuous repeat (rather than one-shot), active low, enabled.

LEDs

The LEDs are controlled via the SL11R's built-in PWM controlled pins (of which there are four, though only three are used). The PWM pins are setup during initialisation (see above).

Colours are sent via the commands (or similar):

with xx being the intensity (or at least the value passed to the command line of wlights).(can probably add more here, we know a lot more about the leds than this now)

Tuning Datastream

The frequency is set by a set of 6 'req=4' commands. The first five always seem to be the same; the last one varies dependant on the frequency. The wtune code just calculates the value for the last command.

Download Radioscape Driverpack

These values are presumably sent the to frequency synth. PLL (LMX2331U); but I've not tried to figure out the actual meanings of the parameters.

I've also been monitoring the 'req=5' commands that are sent by the tune_thread.e.g.cmd=5, value=007f, index=7fff ffff ffff fff8 0000 0000 0000 0000 0000 0000 ff00 056c 0011 0000 000f 0000

The 1st 4 words (0x7fff ffff ffff ffff8) may control which symbols are transferred, but normally I see these values. Sometimes, if the tuner cannot lock (bad freq, or no reception), I see something like (1111 1111 1111 1111) which possibly resets the null symbol synchroniser.The last but few words (056c, 0011) are probably the AFC. If there is no reception or lock then these values are constant. If it is tuned, then they vary slowly, presumably causing the AFC circuit to adjust the frequency to hold the tuner in lock.- onward and upwards ...

The wavefinder return a block of 0x20c bytes per symbol received. Not all symbols per frame are passed accross the USB interface - only those selected by a 'req=5' command (precise format of this command is still TBD).

Automatic Frequency Control

Download Radioscape Driver Download

The biggest missing bit is the AFC control. I figured out how to send the required frequency to the frequency synthesiser, but to complete the tuning, the frequency must be phase locked fairly precisely in order to ensure that each of the carriers end up in the middle of the FFT bin. This is a 2 stage process.The acquisistion phase achieves a course lock, by performing a sequence of correlations of the received phase reference symbol against a locally constructed phase reference symbool. By sliding the local symbol, and performing a correlation for each position, a course offset value can be determined. I can do the correlation, but I don't yet know how to feed the result back to the wavefinder. The acquisition phase can reduce the phase offset (too many phases here ...) to within 1 carrier (1 FFT bin). Once this phase offset has been reduced, the system enters a tracking phase, to provide fine adjustment. This seems to invlove a further correlation, and then uses a moving average to further adjust the frequecy. This process should be able to track any frequency drift, and reduce the phase error to < 1 carrier - i.e centre the sample point to the middle of the carrier. As with the acquisition, I do not as yet have a complete understanding of how the analysis is converted to the value that is sent back to the wavefinder.

Main datastream - processing data blocks

My current thought is that the DSP may also do the demodulation as well as the FFT. This is mainly based on the amount of data that is returned. For each symbol, we get back 1 block of data (0x200) bytes, of which the last 0x80 bytes is a replica of the first 0x80 bytes of a previous symbol - this must be the effect of the guard band.That leaves 0x180 bytes per symbol. A symbol is based on 1536 (0x600) carriers; so that only allows 2 bits per carrier. So, either the data is 2 bits per fft bin, or the fft output has been further processed.The header for each data block indicates which symbol it is for. Symbol 0 (the NULL symbol) seems to be returned for all frames. Sometimes I also get 0x12-0x3d; somtimes I only get 2,3,4.2,3,4 are the FIC; higher numbers are the MSC.

The data returned appears to consist of a 12 byte header that identifies the symbol number, then 0x180 bytes of 'useful' data, then a further 0x80 bytes of 'useless' data (which is presumably due to the guard band).

Download Radioscape Driver Printer

The 0x180 bytes of symbol data is the result of some processing by the DSPs.
The DSPs take in data from the ADC, perform a 2048 point FFT, then perform differential demodulation. The data transfered over the USB is the resulting OFDM complex symbols.
These are transferred as 16 bit words. Care needs to be take to process this data in the correct order - aren't big/little endian issues a right pain...

The symbol data must then be processed accroding to EN 300-401. These steps are:

Frequency Deinterleaving - Each of the 1536 QPSK complex symbols is transmitted on a different carrier. Theses are reordered before transmission. This process needs to be reversed.
Symbol Demapping - The 1536 QPSK symbols are derived from 3072 OFDM symbols. This process is reversed here.
Depuncturing - Extra zero's are added to the data stream according to the puncturing pattern.
Viterbi Decoding - The Viterbi decoder attempts to correct any errors in the data stream, using the redundancy that was added before transmission. Approx. 3 x more data is transmitted than is required.
Descrambling - The data is scrambled to produce a similar number of 1's and 0's.

Command list

(perhaps we should put together a full list of USB commands and packets and what they do)

Patriotfox684