FEC and 4-FSK - modem-concepts for
Highspeed-Fullduplex-Linkstrecken

Alexander Kurpiers, DL8AAU@DB0ZDF, Ludwigshöhstr. 77, D-64285 Darmstadt
email: [email protected]

Summary:

The quasi-standard for Linkstrecken is G3RUH-kompatible FSK-Modulations at presents. With this binary FSK 19,2kbit/s is reached on 1,2GHz. Here, at experimental modem is introduced that with four-stage FSK and modern DSP-Technik with same spread reaches at least the double transfer-installment with comparable sturgeon-spacing.
On the higher microwave-ribbons, spread is existing sufficiently in order more certainly to do the transference with encoding. At modem based on at FPGA is described that with one RS-(15,11) as well as RS-(15,13) block-code until at leases 1Mbit/s can transferences.

High-speed Linkstrecken - to what?

FSK-Modulation now becomes use in the amateur-radio for the realization of the Linkstrecken for years and becomes generally accepted therefore with the users meanwhile. However therefore the demand climbs with the constantly climbing transfer-capacity of the PR-network: WWW and TCP/IPS and therefore digitale language-transfer necessitate always faster Linkverbindungen. Limiting, that effect it to the one at the moment, canal-calculators would use - to the other however the low spread, that is allotted the operating fashion PR. Primarily, the Linkstreckens ary driven on 1,2GHz at presents. With him/it there usual 50kHz-Kanalraster ises at maximum transfer-speed of 19,2kbit/s möglich1 with binary FSKS anuses G3RUH. ¹The transfer-speed has through application of at four-stage FSK doubled with seed spread. Essentially more is with that at present, FM-Transceiverns would of necessary certainly use possibly. One would use New Transceiver, that realizes linearly modulation-procedures like PSK or QAM with I/Q-Modulatoren, would be certainly therefore 76,8kbit/s in 50kHz spreads or more possible. In this article, the Main focus reads can be used on the modem-technology, that usually existing Transceiver shall, without alteration however. The ridge part devotes Time to the realization of such at modem on him/it `` EZKIT lite'' of analogously Devices, at demo-board with the very efficient Festkomma-DSP ADSP2181.

On the higher ribbons (5,6GHz and especially 10GHz) essentially stand's more spread to the disposal. Package-repetitions incriminate the connected knot-calculator strongly, according to that it appears favorably, through encoding and mistake-correction already before the canal-calculator bit-mistakes, that originate on the canal through flush, to avoid. This is to be reached of course only through additional redundancy in the signal. More spread is required for the encoded signal, ace for the uncodierte, what one can with the spread standing to the disposal verschmerzen however. Is now for it through bit-mistakes packages destroyed much more rarely.
At realization of such at modem on at DSP fails because of her/it many too slight pace of thesis processors. That today almost exclusively would use G3RUH-kompatiblen modems of DF9IC or DK7WJ programmable Logikbausteiles uses (GALs). For at similar solution something therefore striven, had to however be chosen at FPGA2 because of the essentially higher complexity of the modem. ²

4-FSK modems for Vollduplex-Linkstrecken

Binary FSK with the known G3RUH-kompatiblen modems does transference practically impossible with more than 19,2kbit/s on the 50kHz wide canals on 23cm. Through more suitable basis-band-filtration or extremely strong reduction of the lifting, another increase is possibly feasible on 28,8kbit/s, more however certainly necessary. The idea to stand-in the transfer-speed is of necessary again [ 3], concrete bases became in [2] described. From thesis bases and New ideas, at modem-realization has originated [ 7], that with at DSP-Board transfer-pace up to 48kbit/s (at leases 38kbit/s) admits.

The modem something optimized to it with calculator-simulation and something implemented anuses it on the EZKIT-lite. It appeared quickly with the calculator-simulations that at efficient modem must be essentially costlier than at modem for 4-FSK for binary FSKS. At hardware-realization with GALs is of necessary possible unfortunately. The necessary Demodulationsalgorithmus can be implemented elegantly by the DSP. Unfortunately, however therefore the maximum step-speed use restricted AD/DA-Wandler with it through the rake-performance of the DSP and above all through them/her/it. Since stand's anyway only 50kHz spreads for Linkstreckens to the disposal, this is no restriction at the moment.

At block-switch-picture of the realized modem is represented in the Bild  1.

  

Illustration 1: 4-FSK-Modem

Basis-band-filtration

With G3RUH-kompatiblen modems, at basis-band-filtration, that takes care of it, is put in that the basis-band-impulse ary band-restricted before the FM-Funkgerät. This procedure is necessary absolutely favorable because of the then following FM-Modulation. If one looks at the phase-impulses of such at filters, according to one recognizes relatively strong fluctuations. Thesis phase-fluctuations cause at higher spread-demand anuses the FM-Modulation however. It is essentially better to make this phase-impulses softer - even if one requires more spread before the modulator with it. Another problem is the demand anuses at opened eye at the recipient. If one admits with the Sendeauge, that neighboring symbol's overlap themselves (`` Partial Response´´), according to spread can clearly be saved. One must correct in the recipient of course then.

If at gaußförmiges filters has proven itself here, with corresponding modulation-index, one comes to the GMSK well known from the mobile-radio.

  

Illustration 2: Impulse-answer and phase-impulses with GMSK BT=0.4

The eye-pattern the here would use Sendefilters is represented in Bild 3. ¹One clearly recognizes that the eye is fully opened for no longer.

  

Illustration 3: Eye-patterns at the modulator-exit

Ace one recognizes in Bild 4, the Sendespektrum is with the chosen basis-band-filtration in approximately immediately broadly like `` herkömmliche'' binary FSK-Modulation.

  

Illustration 4: Ausgangsspektrum with FSK-Modulation

Synchronization

With binary FSK with Scrambler, at bar-synchronization is necessary in the recipient merely. At simple DPLL, that examines the zero-passageways of the reception-signal, is enough to it. By the Scrambler, one tries to do the data median value-freely, according to that at decision between sent `` 1 '' and `` 0 '' very simple becomes: the threshold is put on the median value, d.h 0. That doesn't go with four-stage FSK of course according to simply, therefore here, one can put at threshold in the middle, however therefore the two other thresholds must be decided. At amplitude-regulation is necessary for it before the Entscheider.

The bar-synchronization could take place in the principle over the zero-passageways of the reception-signal again, however is statistical much more unlikely zero-passageways with at four-stage signal, at regulation therefore difficult. In the modem here is used at sogennanter Maximum-Liklihood-Taktphasenschätzer. The temporal drainage of the at the beginning of-signal definitely turns into it (through at corresponding filter realizes) and with the Ausgangssignal of the reception-filter multiplies. This signal becomes gemittelt and uses for the adaptation of the optimally Abtastzeitpunkts.

With him/it here would use AD-Wandler another problem quietly exists: the Abtastzeitpunkt is pretended by the hardware and cannot be adjusted. The trans-molder delivers 2 Abtastwerteses per symbol. Since the Abtastzeitpunkt is aground and should of necessary agree with the optimally Abtastzeitpunkt of course, definitely turns with at so-called Polyphasen-Filter of the Abtastwert into the optimally Abtastzeitpunkt through interpolation. This necessitates no increase-expenditure in the Demodulator, must merely for all possible Abtastzeitpunktes (here 32) filter's existing is. The interpolation is enforced according to that uses the convenient strainer ace reception - as well as drainage-filters becomes.

Correction with DFE

Sendefilter would use through that, the eye is of necessary fully opened at the recipient. By at linear Entzerrer, one can take care of it again however that no Intersymbolinterferenz appears. Mistakes with the Taktrückgewinung therefore cause at Intersymbolinterferenz, that one can therefore remove here. Prerequisite for it is however that the Entzerrer work with Überabtastung - here no problem, since 2 Abtastwerteses per symbol ary processed in the reception-filter anyway. One names such at Entzerrer therefore Fractionally Spaced Equalizer (FSE) [ 5].

The Sendefilter is in the principle `` to schmal''. About the Intersymbolinterferenz to eleminieren, high frequencies must be lifted strongly by the linearly Entzerrer. This leads to at reinforcement of the flush in this frequency-area of course. The flush is colored by the FM-Demodulation however: it accepts to high frequencieses zu ³. One reaches at clear improvement through at additional necessary-linearly Entzerrer. It turns into it, ace shown in Bild 1, the signal behind the Entscheider back-connected. The trick exists with it in it that the difference between the signal before and behind the Entscheider of exactly the disturbance corresponds to that can be withdrawn over the feedback of the signal then [4].

Together with the linearly Entzerrer at the of entrance Mark this very efficient Entzerrerstruktur ace Decision Feedback Entzerrer one (DFE). According to that the Entzerrer work optimally, his/its filter-coefficients ary adjusted adaptive. This needs his/its Time of course, according to that such at signal-processing does only for Vollduplex-Übertragung scythe.

At serious disadvantage of the DFE should of necessary be withheld: since decisions ary back-connected, mistake-reproduction occurs with wrong decisions. That has at effect therefore on the filter-coefficients, that ary adjusted then, then. Snifters fading-break-ins can lead interruptions according to too long, until the modem has put in the filter-coefficients optimally again.

  

Illustration 5: Eye-patterns at the Entscheider with (right) and without DFE (left) with verrauschtem at the beginning of-signal (seed S/N)

One can be able to tell the effectiveness of the DFE by the eye-scrutinizing in Bild  5. Without DFE, the eye is closed practically by flush, with DFE, another quite big opening is recognizable.

Simulation-results

Since no suitable test-surroundings something available during the modem-development, became the essential parts with `` Ptolemy'' ⁴ simulateses. The FM-Sender and -Empfänger something modeled with realistic of filter. In the modulator and Demodulator, the seed of filter-prototype became ace in the DSP-Implementierung uses. At adaptive DFE therefore went it into action, however the bar-recovery didn't become with simulated. The results of the simulation can be read off in the Bild 6: the modem would act anuses this simulation towards the known G3RUH-Modems, that became mitsimulierts ace reference, ⁵ comparably. That means that the transfer-speed of 23cm-Interlinks could be doubleds only by exchange of the modem.

What the simulation has however therefore shown, at essentially more distinctive sensitivity is opposite Frequenzversatz between stations and recipients - in the practice quite well at `` schiefen'' eye, to recognize, - and of course against necessary-linearly distortions. To what extent LinkTRXs would usually use that with that at problem is, at attempt, show wants in the practice.

  

Illustration: Bit-mistake-frequency with flush

Realization

The modem is realized ace program on the Ezkit lite of analogously Devices. On this board, DSPS ary already, Eprom and AD/DA-Wandler existing. Unfortunately, one must quietly modify something at the trans-molder in order to guarantee at sufficient low low border-frequency. The connection at the HDLC-Port takes place over at Steckleiste. There, therefore Status-LEDses ary connected, that etc it shows, whether the DFE adapts again (for example anuses at fading-break-in). At present, there ary two versions of the modem: for 50kHz wide canals and for 25kHz wide canalses. With the second version, 24kbit/s, that ridge does, ary attainable 48kbit/s at present of course only approximately. that could quietly be something too much, can be reduced by at simple program-alteration however (44kbit/s as well as 22kbit/s).

Outlook

The Demodulation with DFE brings clear advantages (approximately 3dB) opposite the simple Entscheider in the G3RUH-Modem. The modem should become widen, according to that therefore binary FSK is possible (perhaps ace `` Notlösung´´, if at Link threatens otherwise to be cancelled). I promise another improvement to me from her/it in [2] introduced encoded modulation.

Very helpful would be 4-FSK therefore for users of course, however the problem of the almost synchronization would have to here ridge (devoid of, amplitude, DFE) is solved. At the moment, the modem is too slow much for it.

FEC modem for Mikrowellenlinks

At mistake-protection is already implemented with Packetradio: the CRC of each package in the final. 2 bytes of redundancy ary attached to it to the data. This is enough in order quite certainly to recognize, whether at package something disturbed by transfer-mistakes. On transfer-canals, where transfer-mistakes ary extremely rare, ace for example cables or light-wave-leaders, this character of mistake-protection is enough. On radio-canals, transfer-mistakes ary much more frequent through thermally flush or other sturgeon-influences, from what one itself through observing that `` Qualität'' of Linkstrecken easily can convince. One joins the figurative package sufficiently additional redundancy (Prüfbits) to it, necessary only one mistake-recognition is according to but therefore at mistake-correction (FEC)⁶ possible.
The encoding-theory delivers us the question, how much Prüfbitses ary necessary and meaningful-prove like the encoding is enforced. It is clear, that - the more Prüfbitses ary added, all the more transfer-mistakes recognized and can be corrected. Therefore the redundancy ascends for it then - instead of Prüfbits been able to become send therefore data. One should manages on preferably little Prüfbits therefore.

In this modem, at Reed-Solomon-Code becomes (RS-Code) uses. RS-Codeses have very good qualities and use therefore frequently. The code is no binary code, I.. the code-symbol's consist from several bits - with him/it here would use RS-(15,11) as well as RS-(15,13) code from in each case 4 bits. At RS-Codewort always abandons itself (2n -1) code-symbol, that consist of n bit in each case, together here therefore from 15 4 bits of code-symbol or 60 bits. In order to can correct at faulty symbol within at code-Word, at leases 2 Prüfsymboleses ary to be intended, one can use from the 15 icons per Code-word therefore only 13 for data. One calls this code RS-(15,13) code. Must analogously for the correction from up to 2 faulty icons at RS-(15,11) code uses becomes. With the correction, it plays no role with it, whether in the faulty symbol only 1 or even finished 4 bits wrong is. Therefore one uses RS-Codes therefore gladly, if one expects bundle-mistakes on the transfer-canal.

Redundancy = reduced net-throughput?

Of course, additional piece of piece of information must be added the too figurative data-stream for the encoding in order to license the recipient at mistake-correction. On the ridge gauze, this goes on expenses of the net-throughput, since can now be transferred few data, ace without encoding. With this contemplation, one doesn't take into account however, that much redundancy is already added with Packetradio by the protocol to the mistake-correction: faulty packages must become repeatedly! If one assumes at Overhead of 20% for at package-repetition, according to one comes to her/it in Bild 7 representeds `` Coderate'' R, i.e. the relation-hip of usefulness-data to figurative data. That here introduced modem uses R=11/15= 0.73 and R=13/15=0.87 alternatively. From approximately 90% goodness, i.e. any 10. Package had to would become repeatedly, the net-throughput is better with the modem with encoding. In the picture, one clearly recognizes that only with encoding with very lower `` Qualität'' additional redundancy through package-repetition necessary becomes.

  

Illustration: Code-installment R ace function of the FlexNet - `` Qualität''

Overview and function-manner

In Bild  8, at simple block-switch-picture of the modem is represented within the transference-route. Through him/it in the modem enthaltenen quartz-oscillator is pretended over at Teiler of the Sendetakt for the calculator. For the Codierer is headed by this pure-signal and is ordered data with at pure-signal of the calculator derived from it. Thesis data become before the encoding in the Scrambler verwürfelt. Behind the Codierer, at pulse-formation of the signal takes place through the Sendefilter. There, there is been for at FM-Sender and is transferred over the transfer-canal.

  

Illustration 8: Block-diagram of the FEC-Modems

The empfangene signal of the radio equipment is ridge filtered in the reception-filter and is cared for at Komparator ace Entscheider. With him/it now digitale Ausgangssignal is enforced at bar-recovery. This pure-signal is used for heading for the decoders. From the pure-signal, the decoders derives another pure-signal, with which anuses the Descrambler, that makes the Verwürfelung, that is relayed data at the calculator, declining.

At systematic Codierer is used, I.. the Prüfbitses ary simply attached to the data. The pure-signal's to the calculator is of necessary continuous, during Prüfbits is sent, no devoid of is generated. So, the adaptation is reached the transfer-guessing.

Realization of the modem

Ace already described, the realization of the modem leans against the successful design of DF9IC (lake [1]) at. So, the station is realized very similarly. Merely, the encoders for him/it would use something supplemented RS-Blockcode and there-transformer and OP's at the higher transfer-speed something adapted. The recipient something clearly widened. The bar-recovery something revised, and should now generate clear less dynamic Jitter. Principal item of the recipient is the decoders for the RS-Code however.

About the modem necessary unnecessarily, to leave, is complicated, all essential function-blocks ary accommodated in at FPGA. The configuration-data of the FPGA ary stored in at EPROM. When switching on the care-tension or anuses causing of at Reset over the modem-interface is configured the FPGA from this EPROM. Simultaneously, this EPROM contains the for the Decodierung and code-word-synchronization necessary tables. The decoders requires long Schieberegister 4 16 bits. Thesis would have been able to become of course therefore within the FPGA encoded, would have covered however precious place there, according to that at bigger and with it clearly teureres FPGA had to would be used. Instead, 2 Schieberegisterses ary put in 4517. At PIC micro-controller is put in additionally for the control of the FPGA configuration and the code-word-synchronization.

The principal item of the modem is the mistake-correction in the decoders. The task of at decoders is general to find the of fake symbol in the code-Word (mistake-position) and, to withdraw the mistake, (mistake-value). At syndromes is normally calculated to it from the empfangenen data. This syndromes is independently from the sent usefulness-data. Normally, at algorithm would now become like Berlekamp-Massey (BMA) uses in order to calculate mistake-position and -wert from the syndromes. This is too costly within the FPGA, therefore at Eprom is addressed with the calculated syndromes and is finished reading the corresponding values. Another trick is applied with it however: with at so-called Meggit-Decoder, the mistake-value only needs to be stored. For the RS-(15,11) code is the syndromes 4 Time 4 bits = 16 bits long, analogously for the RS-(15,13) code 8 bits. AT EPROM gets by with it with 128 kBs for both codes. Additionally to the mistake-value (4 bits) is stored in the remaining 4 bits, how big the mistake-weight, I.. the number of the mistakes in the code-Word, what and whether the mistakes ary because of the start or final of the code-Word. The code-word-synchronization-algorithm uses this piece of piece of information in the PIC. To it becomes in the Eingangsbitstrom ace long ace faded bits, until the mistake-number at while as well as is. $\le 2$$\le 1$ Anuses short Time (little ms with 1Mbit/s) finds the modem the correct start of of the code-Word without additional synchronization-piece of piece of information so. For at Burstübertragung (Halbduplex) is this simple algorithm of course inappropriate, but the modem has specifically been developed for Vollduplex.

Outlook

In the prototype of the modem, at FPGA is used XC3042-PG84-6 of Xilinxes. This component is relatively expensive, according to that is tried to take one in future from stattdessen XC5202. The solution with two Epromses (one for code and FPGA-Konfiguration, that second for the Sendefilter) could quietly be simplified, if the Sendefilter with in the FPGA would be realized. Therefore, that is at present, PIC would use 16c56 too big, according to that wants be used the known 16c84. The modem wants probably be described in the entgültigen version in the coming Adacom-Magazin. Then, there wants be switch-plan and layout to it therefore.

Summary

The introduced modems of solve two problem, the faster Linkstrecken until now prevented,: with 4-FSK, one can finally transference more quickly therefore on 23cm, with FEC, very almost Mikrowellenlinkses become realistic. The modems ary quietly however both in the Prototypenstatus and ridge more detailed tests wants show, whether the concepts prove themselves so. Nevertheless, I wants try to publish switch-plans and circuit board-layouts up to the next Adacom-Magazin therefore. Nearer piece of information becomes in the Hamweb under http://db0ais.ampr.org/~dl8aau as well as in the Internet under http://www.afthd.tu-darmstadt.de/~dl8aau, is found.

Completing, I would like to thank Joachim Erbs and Alexander Ruff that has achieved the essential part at the development of the modems in their work.

Literature

1

Rech, W. -H., (DF9IC): Modern FSK-Modem - compatible to the standard anuses G3RUH. ADACOM magazines 2, S. 13-31, 1991.

2

Kurpiers, A. (DL8AAU): Multi-stage modulation-procedures and her/its/their application in Packet-Radio. Skriptum 12. Internationally Packet-Radio convention, Darmstadt 1996, S. 87-96,

3

Rech, W. -H., (DF9IC): Confessed and New modulation-procedures for pack radio, Skriptum to the 40.+VHF-convention Weinheim, S. 305-329, 1995.

4

Spinnler, B., Huber, J. B.: Aufwandsgünstige incoherent CPM-Übertragungsverfahren with avoidance of mistake-reproduction through signal-formation. Kleinheubacher convention 1995.

5

Bingham, J.,: The theory and practice of modem design. John Wiley & Sons, 1988.

6

Erbs, J.,: Development and construction of at modem to the mistake-secured radio-data-transfer. TH Darmstadt ST 3061 ÜT, 1997.

7

Ruff, A.,: Development and realization of at modem for at 4-FSK radio-transference-runs on at DSP-System. TU Darmstadt D 3093 ÜT, 1998.

Footnotes

... Possible¹

isolated turns therefore 28,8kbit/s or even 38,4kbit/s done. This goes however only with strongly reduced lifting and with it only with very good S/N

... FPGA²

FPGA Field programmable gate array

... Zu³

Therefore one uses therefore Pre - and Deemphase with analogous FM, I.. higher frequencies in the signal ary lifted on the station-side and on the recipient-side again abgedämpft, all the the flush, to can reduce.

... `` Ptolemy''⁴

`` Ptolemy'' is at free simulation-package for Linux as well as Unix

... wurden⁵

The S/N is served ace Eb/N0, covered on seed Bitrate therefore. Since the stand-in Bitrate is reached with 4-FSK with seeds spread however, one must quietly add to the curve 3dB - then is quietly 4-FSK somewhat better however

... (FEC)⁶

FEC forward error correction