1.2MBit/s SuperVozelj QRV
Contents:
1. Hello
2. History
3. Super Vozelj node
4. 1.2Mbit/s PSK RTX
5. WBFM RTX for 70cm and 23cm
6. Plans
7. Acknowledgements
8. Information Sources
9. Links to PSK RTX pages
Hello,
a few days ago, on 12/6/1995, we (S53SM, S59AW, S57MSL and
S53MV) upgraded our first SuperVozelj packet-radio node
GORICA:S55YNG with one port operating at 1.2288Mbit/s with
a PSK transceiver on 2360MHz. The new high-speed node is
currenly linked at 1.2288Mbit/s only to the experimental
SuperVozelj node RAFUT:S59DAY installed at my QTH (S53MV).
Since the new node is certainly the fastest AX.25 amateur-radio
packet node in Slovenia and I belive probably one of the
fastest packet-radio nodes in Europe, I hope that the following
description of our packet-radio node system SuperVozelj will
be of interest to many readers.
Short history of the SuperVozelj project
The developement of faster packet radio links started in
Slovenia already in 1988, when it was clear that 1200bit/s
AFSK Bell-202 links can not cope with the amount of both local
and international traffic crossing our country. In 1989 we
installed the first links operating in the 23cm amateur-radio
band using wideband (250kHz) FM transceivers and simple
Manchester modems at an operating speed of 38.4kbit/s.
At that time the only known choice for node computers was the
TNC2 running NETROM or TheNet software. In the following years
the network of 38.4kbit/s nodes was extended to cover the
whole Slovenia and a few links to some neighbouring countries
were also installed. As the network was growing, it soon became
apparent that the TNC2 and in particular the software was not
able to cope with the traffic at 38.4kbit/s. The TheNet
software was crashing at an increasing rate while new versions
of this software only included new bugs, with none of old, well
known bugs ever being removed.
Therefore a replacement was sought for the obsolete TNC2 and
the poorly written NETROM/TheNet software. Since we could not
find any replacement that could offer something more than what
we already had, in 1992 I started developing our own network
node computer named SuperVozelj. SuperVozelj is based on the
MC68010 CPU and Z8530 SCC chips. At the end of 1992 we already
installed the first operational SuperVozelj node offering six
interrupt-serviced channels for operating speeds up to
76.8kbit/s per channel.
Since the SuperVozelj node computer offered a great improvement
of the network capacity compared to TNC2/TneNet using the same
modems and transceivers, it also made sense to upgrade the user
equipment from 1200bit/s Bell-202 or 2400bit/s Manchester to
higher operating speeds. In 1993 I developed and tested simple
wideband (250kHz) FM transceivers for the 70cm amateur band.
In 1994 several packet-radio nodes were upgraded to have a
192.kbit/s, 38.4kbit/s or even 76.8kbit/s user access in the
70cm band. Since 70cm WBFM radios are cheap and easy to build,
several users followed the upgrade of the network and today the
majority of active packet-radio users in Slovenia operate at
19.2kbit/s or faster.
User upgrades soon caused the congestion of the 38.4kbit/s
interlinks on both 70cm and 23cm. The move to even higher
speeds requires using even higher carrier frequencies with
wider bandwidths. At higher frequencies the avaliable TX power
is getting lower, so more power-efficient modulation techniques
need to be used. My current choice is to use the 2.3GHz amateur
band with PSK modulation at 1.2288Mbit/s. PSK modulation allows
at 1.2Mbit/s a similar radio range as WBFM radios with
Manchester modems at 38.4kbit/s, in other words about 100km
free space range with 20dB link margin and moderate-size
antennas (15-20dBi), as required for typical amateur-quality
packet-radio links.
The developement of suitable 2360MHz PSK transceivers
started already at the end of 1993. The first operating
transceivers were built in the beginning of 1995, followed
by the developement of a DMA-SCC interface card and relative
software for the SuperVozelj node computer. In the beginning
of May 1995 the first QSO was made on 2360MHz, 1.2288Mbit/s
at a distance of only a few meters in my lab at home. On
12/6/1995 we installed the first operational node GORICA:S55YNG
and the first 1.2288Mbit/s QSO was made on a real distance
(6km) using 16dBi antennas. 6km may not seem a large distance,
but there is no optical visibility between the antennas of
the two nodes S55YNG and S59DAY linked at 1.2288Mbit/s, the
obstacle in between (hill) being larger than the 10th Fresnel
zone at a wavelength of 13cm. This simply means that
1.2288Mbit/s packet radio will not bi limited to line-of-sight
propagation of radio waves if reasonably designed equipment is
being used.
SuperVozelj packet-radio node computer
A packet-radio node computer is usually installed on a remote
site that is not easily accessible, generally a mountain top.
Therefore the node computer has to be very reliable and for
this reason the use of IBM PC based hardware and/or software
has to be STRICTLY AVOIDED. For SuperVozelj I decided to use
some modules of my own computer design based on the Motorola
MC68010 16-bit microprocessor.
The latter design includes a
bus with "eurocard" connectors, nonvolatile CMOS RAMs, a
real-time clock and a parallel port.
The Z8530 SCC chip was chosen as the serial interface mainly
because of its easy availability and relative completness,
since it includes a DPLL for RX clock recovery and can generate
both interrupts and DMA requests. On the other hand the Z8530
chip has several hardware bugs that the manual calls "features"
when operating in the HDLC X.25 mode and the bus interface is
only 8 bits wide. A final advantage of the Z8530 is that its
bugs are well known since the chip is widely used.
Right at the beginning I decided to write the SuperVozelj
software entirely in 68k assembly language. The Motorola 68k
assembler offers enough instructions that it is really not
necessary nor convenient to use a higher-level language
compiler for the described application. Writing the whole
software in assembly language also means to have a full
access to the machine code, which is of utmost importance
when writing time-critical applications like a packet-radio
node computer. Thinking about future hardware upgrades, the 68k
assembler is not a limitation, since all practical alternatives
use either 68k microprocessors (MC68020 and follow-ons) or
serial communication chips based on the 68k series (MC68302).
The first practical implementation of the SuperVozelj uses
three Z8530 SCC chips generating interrupt requests directly
to the MC68010 microprocessor without any DMA circuit. The
operating speeds are limited by the interrupt-service routines
to about 76.8kbit/s per channel or 200kbit/s for the whole
SuperVozelj. In this case the internal baud-rate generators
and DPLLs of the SCC chips can be used to significantly reduce
the amount of additional hardware.
The most recent hardware upgrade is the DMA-SCC card that
includes a 4-channel MC68450 DMA chip and a Z8530 SCC chip.
The 4-channel DMA services separately the DMA requests of the
two Z8530 receivers and two transmitters, allowing two
full-duplex channels. The speed limit is set by the Z8530
chip to about 2Mbit/s per channel, but the speed limit of
the 68k bus is also close to this value. At megabit-per-second
speeds the internal baud-rate generators and DPLLs of the Z8530
chip can no longer be used and external clock sources and bit
synchronization circuits are required.
Finally, the SuperVozelj node computer also includes a remote
reset telecommand with a hardware sequence detector. The latter
has never been used in the case of GORICA:S55YNG, since the
SuperVozelj software NEVER CRASHED in two years and a half of
continuous (battery-backed) opration. The remote reset circuit
is therefore only used to correct for sysop mistakes, since
the SuperVozelj system allows remote loading and testing of new
software versions, stored conveniently in the nonvolatile
CMOS RAM.
1.2Mbit/s PSK link transceiver for 2360MHz
The choice of a transceiver design for high-speed packet is not
simple. Is it better to use an apparently simplier FM
transceiver or to go for a more sophisticated PSK transceiver?
Both choices have their advantages and disadvantages and at
this time it is difficult to predict which will become more
practical. Personally I decided for PSK transceivers since
there are more different designs available and upgrades
possible than with FM transceivers and the radio range is
between 5dB and 15dB larger than with FM transceivers.
In packet radio the main problem of a PSK transceiver is the
initial RX signal acquisition. The latter is a function of the
carrier frequency uncertainty. In a simple biphase PSK system
with 0/180 degrees modulation, the initial signal acquisition
requires a complicated searching loop, if the frequency error
exceeds 10% of the bit rate. Therefore PSK becomes simple if
you go fast. On the other hand, making amateur packet-radio
satellites to run at only 1200bit/s makes the initial signal
acquisition extremly difficult and this is perhaps the best
known example of wrong PSK design.
Considering the 13cm amateur band, the sum of the frequency
uncertainties of both receiver and transmitter is at least
10kHz using top quality temperature-compensated xtal
oscillators. A real-world figure is 100kHz frequency
uncertainty that requires a MINIMUM bit rate of about 1Mbit/s!
With the above restriction a convenient choice is to use
1.2288Mbit/s. This figure can easily be obtained with standard
baud-rate xtals, being the 32nd multiple of 38.4kbit/s or the
1024th multiple of 1200bit/s.
As already mentioned, a PSK tansceiver allows many different
designs that have to be investigated. In the first four
prototypes I obtain direct PSK modulation of the transmitter
in a balanced mixer operating at the final frequency 2.36GHz.
The receiver uses double downconversion to 75MHz and then to
10MHz, where the PSK demodulation is performed. The RF front
end for 2.36GHz uses relatively new SIEMENS semiconductors
in SMD packages, originally developed for cellular telephones
operating at 900MHz and 1.8GHz.
The transmitter includes a multiplier chain starting from a
18.4MHz fundamental resonance xtal. The same BFX89 oscillator
transistor is used as a multiplier by 4 to obtain 73.7MHz. The
multiplier chain then follows with another BFX89 (147.5MHz),
yet another BFX89 (295MHz), a BFR91 (590MHz/+10dBm) and a
BFR96 (1180MHz/+10dBm). The 1180MHz carrier is applied to a
harmonic balanced mixer with a quad schottky diode BAT14-099R
that produces about -5dBm of PSK modulated signal at 2360MHz
after the necessary filtering.
The PSK signal at 2360MHz is first amplified by a CFY30 to
about +12dBm and then by a CLY2 to about +27dBm. The antenna
switch uses a series PIN diode BAR63-03W and a shunt PIN
diode BAR80. The reciever uses a CFY35 as the first RF amp,
followed by a CFY30 in the second stage. The first
downconversion is performed by a BAT14-099 double schottky
diode harmonic mixer, fed by a LO PLL synthesizer producing
1147.5MHz.
The IF chain includes two amplifier stages at 75MHz (BFR90 and
BF981), a second mixer to 10MHz (BF981) and an IF amplifier/
/limiter (CA3189). The PSK demodulator is a squaring-loop
type demodulator. The main demodulator functions are performed
by exclusive-OR gates (74HC86) used as mixers and frequency
multipliers. The VCO uses two 2N2369 transistors and a 74F74
(dual D-flip-flop) frequency divider.
All of the RF modules of the 2.36GHz transceiver are placed in
shielded boxes made of 0.5mm thick brass plate. The shape of
these boxes is rather elongated (40mm width) to avoid internal
resonances. There are in total 7 (seven) such shielding boxes
in one transceiver, therefore the assembly of such a radio
requires a lot of work and patience. The only unshielded
module is the SuperVozelj interface. The digital signals are
kept at TTL levels, but the inputs and outputs have 75-ohm
terminations in the case the transceiver is installed close
to the antenna to avoid microwave cable losses.
The 1.2288Mbit/s transceiver also requires an interface at the
SuperVozelj side, interface that used to be called "modem"
at lower bit rates. The interface includes 75-ohm drivers
and receivers, a bit-synchronization DPLL and a scrambler/
/descrambler. The DPLL design is an interpolation DPLL that
only requires a 8-times higher clock frequency (9.8304MHz)
using conventional 74HCxxx logic, although provides the
resolution of a /256 conventional DPLL with a 315MHz clock.
The scrambler/descrambler was added for two reasons: to provide
a smooth link degradation in the case of multipath propagation
and to drop the DC coupling requirement that complicates both
FM and PSK transceiver design.
User access WBFM transceivers for 70cm and 23cm
The original idea of the wideband FM transceiver is to have
simple and cheap radios for fast packet-radio links. Wideband
here means a bandwidth of 200-250kHz, since one can easily get
ceramic filters for different center frequencies that all have
the same abovementioned bandwidth. The first WBFM radios for
23cm were put into service in Slovenia in 1989 together with
simple Manchester modems operating at 38.4kbit/s.
Since our WBFM radios were installed before the K9NG/G3RUH
modems became known to a wider audience, we never considered
using K9NG/G3RUH FSK modems in our network. Our only experience
with K9NG/G3RUH FSK modems is that they require very critical
modifications to the audio and PLL circuits of expensive
commercial transceivers and that the final result is uncertain
even on the same transceiver model from the same manufacturer.
Since the WBFM radios proved to be cheap and reliable, we
never considered installing a more expensive system that is
known to have inferior performances.
Encouraged with the success of the 23cm WBFM radios we tried
to develop simple WBFM radios for user access in the 70cm
band. Here one could argue that WBFM radios occupy too much
precious bandwidth in the 70cm band. In practice we found that
the opposite is true: WBFM radios with Manchester modems
actually produce LESS INTERFERENCE to other users than NBFM
radios with Bell-202 or G3RUH modems. The TX power of a WBFM
radio is spread over a much wider bandwidth, so the power
density is much lower. A WBFM radio will never open the
squelch circuit of a NBFM voice transceiver, so FM voice
users will seldom even notice the exsistance of a WBFM packet
digipeater operating on the same frequency they are using for
voice contacts. In other words, WBFM radios represent a simple
form of spread-spectrum communications and spread-spectrum
systems are usually proposed to limit mutual interferences.
Both 23cm and 70cm WBFM radios are similar in their basic
design. Both radios have a PLL controlled TX with an output
power between 1-2W and a receiver with a double downconversion
to a final second IF frequency of 10.7MHz or 6MHz. The 23cm
version is complicated because of the obsolete design, using
some modules of a 23cm linear transverter. On the other hand,
the 70cm version was made as simple as possible to be easily
reproducible (two printed circuit boards of 60x120mm each).
Manchester coding is sometimes also referred as biphase coding.
Each bit is split in two halves, the polarity of the second
half being opposite to the first half. In this way the DC and
low frequency components are eliminated, simplifying the
interface of a manchester modem to a FM transceiver. A
Manchester modem can operate at up to 2400bit/s through an
unmodified NBFM voice transceiver, at 4800bit/s through a
NBFM transceiver modified for the G3RUH modem and at up to
76.8kbit/s through a cheap WBFM transceiver. A Manchester
modem is very simple, about 10 pieces of 74HCxxx chips on a
75x100mm single-sided PCB including the DCD circuit and analog
interfaces.
Plans for future upgrades
The above description only describes our first experiments at
1.2288Mbit/s. Although the described hardware is currently
operating as expected, it still has several drawbacks. The
SuperVozelj node computer itself is unable to handle the full
capacity of two 1.2288Mbit/s channels. Therefore the CPU will
have to be replaced with a MC68020 or even faster
microprocessor if the full channel capacity is to be used.
The interface between the MC68450 and Z8530 and the Z8530
itself is not the best choice for speeds above 1Mbit/s.
A MC68302 serial controller would be a much better choice.
The described PSK transceiver is both complicated and slow,
requiring a TX delay of 2-3ms that is fairly excessive at
1.2288Mbit/s. This delay is mainly caused by the turn-on
transient of the transmitter oscillator. PSK reception also
allows a direct-conversion receiver just like SSB. The
receiver and transmitter of such a direct-conversion
transceiver could have many stages in common and no critical
IF at all, while the TX delay would no longer be limited by
xtal oscillators.
Moving to higher speeds we radioamateurs will also have the
opportunity to explore different
antenna position, polarization
or other diversity techniques to improve the reliability of
contacts affected by multipath or other propagation anomalies.
High-speed data transmission in these conditions and at these
data rates is relatively new also to professionals, so we can
really contribute something new to the propagation science.
Acknowledgements
The SuperVozelj project is not just my own work. It also
represents the work of a large number of radio amateurs and
here I am only going to acknowledge the most important
contributions:
Knut Brenndoerfer, DF8CA, helped me obtaining state-of-art
semiconductors for the 13cm PSK transceiver. Without his help
the transceiver would be even more complicated using obsolete
parts.
Mijo Kovacevic, S51KQ, and Tomi Kacin, S57BKC, designed some of
the printed circuit boards of the SuperVozelj computer and
modems.
Iztok Saje, S52D, made several upgrades to my original software
to make it compatible with different BBS, DXCLUSTER and other
packet-radio node programs.
Franci Mermal, S51RM, and Sine Mermal, S53RM, developed new,
more compact printed circuit boards for the SuperVozelj and
for the 70cm WBFM transceiver.
Information sources
All of the information about the SuperVozelj is freely
available. Most of the hardware has already been published
in our national Slovenian amateur-radio magazine "CQ ZRS".
The 13cm PSK transceiver will be published soon, probably
in the next issues.
The SuperVozelj source code programs and instructions are
regularly loaded on LJUBBS:S50BOX under the directory DSP3MV.
However, most of the instructions and comments to the source
code are only avaliable in Slovenian language.
Finally, the SuperVozelj node and WBFM transceivers were on
exhibit last year in Friedrichshafen (Germany) on our
organization ZRS stand. This year we hope to show the new
developements (13cm PSK RTX, new versions of the WBFM RTX
and SuperVozelj node computer) in the same place (ZRS
stand in Friedrichshafen). I am also planning to prepare a
lecture on our SuperVozelj system for the VHF-Meeting in
Weinheim (Germany) to be held on September 16/17, 1995.
73s de Matjaz Vidmar, S53MV @ S50BOX.SVN.EU
Links to PSK RTX pages
23cm PSK RTX.
13cm PSK RTX.
An improved
BPSK demodulator for the 1.2Mbit/s
packet-radio RTX (by S53MV, Matjaz).
Latest redesign of the PSK RTX was made by S51RM and S53RM;
new design means also some improvements over basic design.
