D-STAR, Part 2 of 3: Design Considerations

Icom America expresses its gratitude to the ARRL for the permission to reprint and post this review on our Website. Copyright ARRL.

Come learn what JARL put into their proposed new VHF communication standard.

By John Gibbs, KC7YXD

n the first segment of this series, the US) that is little used today. In fact, directional, asynchronous nature of we considered the attributes we we should use this spectrum or risk los- multiple contacts demands a full- Iwould like to see in any new VHF/ ing it to commercial interests. Also, if duplex repeater link. UHF Amateur Radio system. In this we want to develop a system to send Given the high-speed data require- second segment, we discuss the tech- high-speed data, we will need a wide- ments of the system and the desire to nical issues involved with selecting the bandwidth signal and there is little use digital voice to reduce transceiver parameters of an Amateur Radio digi- available spectrum at 70 cm. If it is de- spectrum requirements, the repeater tal system to meet those attributes. sired to use the D-STAR protocol at link should be digital. With the asyn- Perhaps the easiest way to organize lower frequencies, the high-speed data chronous nature of the system, packet the design considerations for a digital mode could be dropped. In fact, a pro- mode is a natural choice. However, radio system is to use the OSI Model. totype portable 2-m HT using only the because a packet system does not The OSI Model is officially known as digital voice mode has been developed guarantee real-time communication, The Basic Reference Model for Open and was shown this spring at IWCE. voice should be given priority over Systems Interconnection. We start the Like the previous FM system, the data to minimize the possibility of description with the bottom level of the system design logically calls for half- voice disruptions. OSI model, the physical layer. Then we duplex operation for digital voice and will work our way up to the top levels, simplex for high-speed data. Modulation which are open to amateur experimen- An ideal modulation system should tation and application development. Repeater Link Frequency generate a signal that has a narrow Since it is desired to have multiple spectrum, with low side lobes so that Physical Layer contacts and high-speed data packets it does not interfere with other nearby Transceiver Frequency on the repeater link, a wide bandwidth users. On our crowded ham bands, this We hams have a large spectrum al- is required. Therefore the repeater is becoming more of an issue daily. location at 1.2 GHz (60 MHz wide in backbone must be at microwave fre- There are several commonly used quencies and the amateur band at modulations for digital data. The digi- 10 GHz is a logical choice. Today’s tal modulation scheme chosen will sig- 18225 69th Pl W equipment generates usable 10 GHz nificantly affect the performance of a Lynnwood, WA 98037 power, is affordable and the 10 MHz communication system. Generally we [email protected] of bandwidth is practical. The bi- want to maximize the data rate within

22 Sept/Oct 2003 Reprinted with permission; Copyright ARRL. the constraints of acceptable level of interfere with the decoding of the de- trum) is also used for multiplexing cel- latency, available bandwidth, accept- sired symbol, a phenomenon called lular phones. In CDMA, several cell able error rate, product costs and op- inter-symbol interference (ISI). phones share the same frequency and erating environment (that is mobile, A typical compromise between ISI transmit simultaneously. Each phone portable, fixed-link). In particular, and bandwidth used by many systems on a frequency is modulated with a code mobile and portable operation causes is for the bandwidth/data rate ratio to sequence that spreads the spectrum in variable multipath fading and fast be equal to 0.5. This yields almost no a unique way. If the receiver is synchro- phase shifts that can wreak havoc degradation due to ISI compared to nized and has the same code sequence, with digital radios. MSK and yet dramatically reduces the then the signal is restored. Otherwise, Less spectrally efficient modulations spectral occupancy of GMSK com- the signals from other phones become generally have better operation charac- pared to MSK. part of the background noise. teristics in poorer SNR conditions. Also, QPSK—In theory, quadrature An important limitation on the sys- they are more forgiving of frequency- phase-shift modulation could have a tem is that undesired signals are not offset errors between the transmitter constant amplitude format. However, completely rejected. Depending on the and receiver and frequency and phase the rapid switching of the input data length of the codes used and the atten- response error on the channel—an causes a QPSK signal to have large dant difficulty in synchronizing, per- important consideration if costs are to sidebands that destroy its spectral effi- haps 20-30 dB of so-called processing be kept down in a UHF system. ciency. Therefore in practice, raised co- gain can be attained. Therefore, a strong 4 FSK—FSK, MSK and GMSK are sine filters are used on input data to nearby CDMA signal can overpower a very attractive because they are con- reduce these sidebands. To preserve the more distant signal. This classic prob- stant amplitude modulation. This wave shape induced by these filters re- lem with spread-spectrum communica- means that the power amplifier can quires the use of a more expensive and tions is called “the near-far problem.” be class C, which offers low cost and less power-efficient linear amplifier. If In a cell-phone system, this problem excellent power efficiency. a class-C amplifier were used with is addressed by power control. Since all FSK has of course been used in QPSK, the sidebands that were the nearby phones are communicating amateur systems for years, dating removed by the cosine filter would be with the same nearby cell site, the cell back at least to the introduction of regenerated. site remotely controls the power level RTTY. Newer variations on FSK use In the presence of additive white of each phone to minimize the possibil- more frequencies than just mark and Gaussian noise (AWGN), QPSK re- ity of interference. However, in Amateur space. For instance, the new weak sig- quires about 3 dB less signal-to-noise Radio, particularly with multiple sim- nal mode, JT44 uses very slow FSK than does FSK. However, in real chan- plex contacts, this is not a solution. (about 5 Hz data rate) with 44 differ- nels, with multipath and poorly ent frequencies each corresponding to synchronized receivers, the 3-dB Frequency Domain Multiplexing a character.1 But at the higher data advantage quickly disappears. (FDM) rates needed for a VHF/UHF digital Data Link Layer TDMA, CDMA and other modern voice system, four FSK frequencies multiplexing schemes require coordi- offers an attractive option for Time Division Multiple Access (TDMA) nation between the units that is in- improved FSK performance. TDMA is one of the two commonly compatible with the basic goals of the GMSK—Among FSK, MSK and used multiplexing standards for cellu- Amateur Radio Service. One of the GMSK, GMSK offers the best spectral lar phones. The cell tower site acts as major justifications for our service in efficiency with only a slight degrada- the master clock and assigns a time slot the US is emergency service. TDMA tion in the BER compared to FSK and to each of several cell phones that are and CDMA require an infrastructure MSK. These advantages have made assigned the same frequency. For proper to provide the coordination. This in- GMSK one of the most popular digi- operation, it is critical that each phone frastructure would quite possibly be tal modulations worldwide. Other transmit and receive exactly in its as- destroyed in an emergency. So, the best more complex modulations like QPSK signed time slot. This is not attractive solution for Amateur Radio is what we require a more expensive linear power for amateur simplex operations because have traditionally used, FDM. amplifier that also typically requires operations are as two or more equals, more current, which is critical in por- and there is no master to determine the table operation. clock and assign time slots. Network Layer GMSK low-pass-filters the data Any Amateur Radio system has to In the network layer, the binary stream with a filter that approximates work without a centralized frequency data stream is divided into discrete a Gaussian time and frequency reference and master clock. In addi- packets of finite length. In addition, response. A Gaussian filter is used tion, the radios must be able to acquire error checking is performed by cyclic because of its desirable properties in signals that are somewhat off fre- redundancy check (CRC) at this level. both the time and frequency domains. quency and acquire timing without the If an error is detected, it is corrected This filtering reduces the high-fre- need for a separately transmitted by the retransmission of packets. quency content of the modulation and clock signal. These requirements may therefore narrows the frequency spec- make an amateur system less spec- Transport Layer trum of the modulated signal while trally efficient than a centrally-con- In the transport layer, we multiplex widening the data response minimally. trolled system like the cell phone, but and split all the data streams we need However, as you continue to narrow they are more in keeping with the to send and receive. In an Amateur the filter, the spectrum continues to spirit of Amateur Radio, particularly Radio system, we would typically need narrow and the time response of the the capability to operate when the to include repeater control data; filter lengthens. This causes the peak infrastructure is destroyed. source, destination and routing infor- amplitude to decrease and the adja- mation (that is, call signs of both op- cent data tails of the time response to Code Division Multiple Access erators and repeaters used); and what (CDMA) is called the payload, which is the voice 1Notes appear on page 28. CDMA (also known as spread spec- or data to be sent.

Sept/Oct 2003 23 Presentation Layer results are averaged together.2 ication paths. As Digital Voice Codec A very important factor in conduct- Systems point out on their Web site, ing MOS tests is the acoustical envi- “Vocoders…designed for extremely As mentioned in the first segment ronment. Since the codec is designed low bit-error rates, such as those en- of this series, simple PCM encoding of to highly compress the information in countered in land-line communica- voice results in a 64,000 bits/s data a human voice, it is easy to imagine tions, often experience serious stream. Codecs have been developed that the presence of other signals and degradation when applied to the much to compress voice with good quality noise can severely affect the perfor- higher bit-error rates found in wire- down to 2400 bits/s and lower. These mance of the system. An excellent test less communications. Consequently, codecs develop their extreme data for Amateur Radio is performance in it is important to consider robustness compression by modeling short seg- an automotive environment including to channel degradations during the ments of the human voice and only engine noise and wind noise from an vocoder-algorithm design process.” transmitting the reduced information open window. needed to describe the voice model. A final issue in codec selection is Scrambling One of the major difficulties in the MOS performance in the presence Bit synchronization and accurate designing a digital voice radio is in test- of the channel impairments we com- level slicing in the receiver require fre- ing the voice quality. High-compression monly find in VHF/UHF commun- quent transitions in the data (no long codecs are designed to work with a human voice; traditional tests like frequency response and harmonic distor- Table 1—D-STAR Transmission Characteristics tion with sinusoidal tones do not generate meaningful results. Consequently, Mode Transmission Speed Bandwidth a subjective method of testing called Backbone 10 Mbps or less 10.5 MHz mean opinion score (MOS) has been Data 128 kbps or less 130 kHz developed. MOS is estimated by a test Digital Voice with a group of normal listeners who ITU 8 kbps 9 kHz are asked their opinion on a five-point scale (1 = bad, 5 = excellent) and the AMBE 2.4 kbps 5 kHz

Fig 1—Header additions with TCP/IP protocol.

24 Sept/Oct 2003 strings of ones or zeros). To ensure that side of the transceiver and work our mode and in the digital voice mode. condition, most digital radio systems way out to the antenna, working our If the standard is approved as pro- use a device called a scrambler to ran- way down the OSI model. We will posed, it will contain the following domize the input data stream. then see how the standard defines information: “Scrambler” is an unfortunate term the repeater operation and the links The first two fields are common to for people who are familiar with ana- between repeaters. First we will con- most digital radios, the bit sync and log radios; it is common to interpret this sider the high-speed data mode and the frame sync. These preambles are as encryption. The FCC of course then the digital voice. designed to allow the receive modem forbids encryption for amateurs. Scram- to establish timing and level lock as bling is not an attempt to hide the mes- High-Speed Data quickly as possible. sage content, however; it is a fixed and The standard interface for high- The flag field describes the content published method known by all poten- speed data into and from the D-Star of the data field. tial receivers for converting the input system is IEEE802.3 (10BaseT Ether- Bit 7 Data or voice communication data stream into a data stream with net.) In Fig 1 you can see how the flag. short strings of ones or zeros. D-STAR transmitter adds a radio Bit 6 Repeater or simplex flag. Scrambling is typically done with header extension to the Ethernet mes- Bit 5 Communication-interruption a shift register and exclusive OR gates. sage just as the Ethernet protocol added flag. CCITT recommendation V.26 recom- a header to the Internet Protocol, TCP/ Bit 4 Control signal, data or voice mends this procedure. IP. Since this radio header is stripped signal flag. off in the receiver, the radio link appears Application Layer Bit 3 Emergency/normal signal to be a “wireless Ethernet cable.” There- flag. This is the layer where hams can fore, it is possible using existing Bits 2-0 Transmission-control begin to customize the system and add software (such as browsers) to commu- bits (see below). their own applications. In addition, nicate the same images, text and voice The ID field can hold four call signs: this is the level where the system de- as is handled by Ethernet, including sign allows for user control entry and links to the Internet, without modi- 1. The local repeater you are access- data entry, both data from the IEEE fication. ing (optional). 10BaseT Ethernet and analog audio 2. The linked distant repeater the from the microphone. Data Multiplexing called party is using (optional). Fig 2 shows the details of a com- 3. The station you are calling (can be D-STAR Proposed Standards munication packet from the radio part CQ). As stated in the first part of this in Fig 1. Each packet consists of a 4. Your own call sign. article, D-STAR is not a finalized stan- radio header and the Ethernet packet The PFCS field is a check word for dard at the time of this writing. How- described above followed by an error- the header. Some of these bits require ever, the field trials are finished and checking frame. The radio header is a little more explanation if we are to standard publication begins here. worthwhile to study in some depth as understand the operation of the sys- Table 1 shows the system as it stands it shows many of the D-STAR system tem. Notice that when bit 3 of the flag at this writing. capabilities. field is set, you are asking for an emer- To describe the proposed D-STAR Each frame of the radio header is gency break-in. (On many FM repeat- standard, we will start at the input identical in both the high-speed data ers you would say “break” today.) For

Fig 2—Proposed bit pattern in high-speed digital mode.

Table 2—Call-Sign Combinations and the System Function (Uses Evergreen Intertie Call sign examples3) Called Departure Destination Own Station Repeater Repeater Station Function CQCQCQ KB7WUK K7NWS KC7YXD CQ Portland N7ABC KB7WUK K7NWS KC7YXD Call N7ABC in Portland N7ABC K7NWS K7NWS KC7YXD Call N7ABC on local repeater N7ABC DIRECT DIRECT KC7YXD Simplex

Sept/Oct 2003 25 instance if you want to report an acci- the two codecs were undergoing field open public standard codec whereas dent, pushing the emergency key on trials; today the JARL has selected AMBE is the patented intellectual the radio will set bit 3 and all D-STAR AMBE as the standard. The two stan- property of Digital Voice Systems. radios within range will open squelch dards under consideration were the ITU Unlike many companies, however, the and their volume to be set high. standard G723.1 and a Digital Voice present owner of this technology sup- The flag field bits 0, 1 and 2 are Systems proprietary codec that uses the ports the Amateur Radio community used for transmission control. They AMBE algorithm. and is willing to sell these parts in implement functions like ACK, ARQ ITU G.723.1 uses an ACELP (alge- small quantities. and repeater control. braic code-excited linear prediction) The JARL is not alone in deciding One of these repeater control func- algorithm that generates a 5.3-kbps on AMBE for its high voice quality and tions is repeater lockout. Repeater data stream. With an algorithm delay very low bit rate. Table 3 shows several lockout is used mainly to block illegal of 37 ms, the total wireless-communi- digital systems that have standardized stations. A D-STAR repeater can hold cation-throughput delay is a little over on this codec technology. For instance, a black list of call signs that have con- 100 ms, quite reasonable for half- the Telecommunications Industry Asso- sistently violated repeater and/or FCC duplex communications. ciation (TIA) selected DSVI’s codec tech- rules. If a blacklisted station calls the AMBE stands for advanced multi- nology over CELP and other codecs for repeater, the repeater does not repeat band excitation. AMBE can use differ- the APCO Project 25 North American the message but instead calls the ent levels of compression to trade off land-mobile radio-communication sys- offending station back with the lockout voice quality and bit rate. Tests show tem. This is particularly significant be- bit set. The offending station’s radio will that at the 2.4-kbps data rate, the voice cause at least two Amateur Radio then display a message indicating that quality was at least as good as the groups are evaluating Project 25 radios it is blocked from the repeater. So now, higher-data-rate ITU G.723.1 over real as an alternative digital radio standard it is not necessary to shut down the radio links. The algorithmic delay is for amateur usage. repeater for everyone when one indi- only slightly longer than G723.1 Fig 3 illustrates the bit pattern used vidual is misusing the repeater. (44 ms), so the factor-of-two improve- in the digital voice mode when the Another important field for under- ment in data rate (and spectral effi- AMBE codec is used. As mentioned standing the capabilities of the system ciency) comes with no noticeable latency before, the radio header is identical to is the ID field. Understanding the ID increase. the high-speed digital mode radio field is important because it shows the The data-rate reduction from the header and so will not be discussed great flexibility available in the sys- AMBE codec is particularly significant here. tem calling capabilities. The first thing because of worldwide pressure from The most interesting part of Fig 3 to notice is that the D-STAR protocol regulatory agencies to reduce the oc- is that the digital-voice data frames automatically IDs at every transmis- cupied bandwidth of voice communi- are interleaved with data frames. sion. This easily meets the FCC ID cations sufficiently to allow 6.25-kHz These frames are currently reserved requirements for ID at start, end and signal spacing. When using a modula- by the D-STAR standard with no dedi- every 10 minutes of transmission. tion scheme sufficiently robust to give cated usage by the system overhead. Next, to understand how the four ID reliable communication in mobile and This means that the system is capable fields work, Table 2 illustrates the con- portable applications, only the AMBE of supporting a 2400-bits/s data tents of each field if KC7YXD were to data rate meets this signal-spacing stream from a user application while transmit on a fictional D-STAR Ever- requirement. the user is talking on the system! green Inter-tie system. It is not neces- The decision between codecs is com- Notice that the D-STAR system itself sary to always fill in all the ID fields. If plicated by the fact that G.723.1 is an provides no error detection for this you respond to a CQ or a call directed data, so it would be up to the user’s at you, your D-STAR transceiver will application to provide error detection automatically fill in the fields for you. Table 3—AMBE Vocoder-Based and error correction. This and other Systems overhead would decrease the end-to- Digital Voice end data rate slightly; but if radios are Codec—Those of you who had a Inmarsat built to exploit this capability, hams chance to see the D-STAR presentation Thuraya could potentially add many interest- at last year’s Dayton Hamvention or at Iridium ing features to the D-STAR system. the DCC in Denver last fall may recall APCO Project 25 (IMBE) What is not shown in Fig 3 is that that the Digital Voice mode occupied G4GUO & G4JNT HF Digital Voice the frame and sync fields are repeated often so that the errors between the 8 kHz of bandwidth using the ITU System G.723.1 Codec standard. At that time, transmitter and receiver clocks can be

Fig 3—Bit pattern in digital-voice mode.

26 Sept/Oct 2003 corrected without requiring a master better performance than GMSK or substantial, particularly with the ex- clock signal. It also means that another 4-FSK and its higher spectral effi- tremes of temperature found in portable amateur can tune into the middle of a ciency is obviously attractive. How- operation. The alternatives are to suf- contact and listen to the conversation ever, QPSK’s higher spectral efficiency fer the expense of a precision frequency without waiting for the sync frames of also leads to higher susceptibility to reference in all the radios or adopt a the radio header at the next over. transmission impairments such as modulation method like GMSK that is multipath and phase hits. Yet, the big- more tolerant of frequency errors. Modulation gest disadvantage of QPSK is the need However, GMSK is not so spectrally Several modulation methods were for extremely linear power amplifiers efficient as QPSK. For instance, at investigated during the development of to avoid spectral growth—what we 128 kbps, GMSK with a BT product of 1 the D-STAR standard. Modulations Amateurs call splatter. /2 occupies a bandwidth of 135 kHz. tested included GMSK, FSK, 4-FSK, The front-runner at the time this For the same data rate, QPSK requires MSK and QPSK. GMSK has been article is written is GMSK. In its favor, only 83 kHz. selected for the backbone line between GMSK is a well-proven technology, and The best solution is probably to use repeaters. The standard for the portable probably the most commonly used digi- a codec with the AMBE algorithm de- and mobile transceivers may include tal modulation in the world for portable scribed earlier that reduces the data more than one modulation format. applications (see Table 4). GMSK has rate as far as possible and then use Gaussian minimum-shift keying two basic advantages. First, it is more GMSK for more robust communica- (GMSK) and quadrature phase-shift robust than QPSK to common transmis- tions, but the tests still continue. keying (QPSK) are the two finalists. A sion impairments. Second, GMSK, as a Fig 4 is a somewhat busy graph third, 4-FSK, has been recently pro- form of FSK, has constant amplitude that dramatically shows the difference posed as an alternate standard and is and can therefore use very efficient in occupied bandwidth. The existing now under investigation. The reason for class-C power amplifiers. Third, GMSK FM system bandwidth can be deter- the delay is that selecting the best is not as sensitive to frequency errors mined by Carson’s Rule to be about modulation for D-STAR real world between the transmitter and receiver. 16 kHz. While not all combinations of applications is not a trivial exercise. In Because no master frequency reference modulation and codec algorithms are real mobile communications systems, is available in the D-STAR system, tun- shown, you can clearly see that you the link between a moving node and a ing errors on a 1.2-GHz signal can be can fit many more digital voice con- base station will be subject to tacts into the same spectrum. multipath, which results in Rayleigh Table 4—GMSK is used in Systems fading. This will have a significant ef- Repeater fect on the resultant BER performance, Worldwide The digital voice mode is half- possibly increasing the required C/N for GSM cell phone duplex with a 20-MHz offset between a specific BER by as much as 10 dB. DECT transmit and receive frequencies. QPSK is commonly used in fixed- Cellular Digital Packet Data (CDPD) High-speed data is simplex. link communication systems. Under As shown in Fig 5, much of the re- Mobiltex ideal conditions, QPSK would give peater site function is to provide a

Fig 4—Occupied bandwidth of digital radio.

Sept/Oct 2003 27 Fig 5—Integrated site with analog and digital radio repeaters, high-speed backbone and Internet connection.

gateway to other repeaters, both at 10-MHz-bandwidth backbone link be- existing Internet Protocol. other sites and to repeaters on differ- tween repeaters. This wide bandwidth Backbone field tests have been car- ent modes. The repeater also provides allows multiple voice and data con- ried out with a 36-dB-gain parabolic an interface to the repeater backbone tacts to occur simultaneously on the antenna and a 1-W transmitter. Heavy link to the Internet if desired. The sys- link. An analysis of the frequency of rains in southern Japan of more than tem is designed to support remote con- use of data and voice communications 12 inches per hour limit the practical trol of the repeater over radio and/or demonstrated that a 10-Mbit/s full- distance that the repeaters can be landline links. duplex link would support the needs separated. It was found that taking into Repeaters could be linked via the of up to 12 linked repeaters. consideration these extreme weather Internet instead of the backbone, but The high-speed data and digital- conditions, the maximum range for because of bandwidth limitations, voice data streams from multiple re- uninterrupted communications is about much of the high-speed multiple con- peaters are multiplexed into a single 12 miles. Obviously, the fog and rain- tact capability would be lost. data stream according to the asynchro- fall at the location and the acceptable nous transfer mode (ATM) standard. probability of communication interrup- Repeater Call Sign Protection This 10-Mbit/s data stream is GMSK tion dramatically affect this number. To protect repeaters from co-chan- modulated onto a 10-GHz carrier, re- nel interference, CTCSS tones are sulting in an approximately 10-MHz Notes used to prevent interfering signals wide signal. 1“JT44: New Digital Mode for Weak Signals,” from triggering the repeater. In the D- The ATM cell is made up of a short (World Above 50 MHz) QST, June 2002, STAR system, the digital header con- 53-byte packet that consists of a 5-byte pp 81-82. 2 tains the call sign of the repeater to header and a 48-byte payload. The ATM D. Smith, KF6DX, “Digital Voice: The Next be accessed. If the repeater does not cell is sent to the required destination New Mode?” QST, January 2002. For a discussion of MOS, see the sidebar, “How see its call sign in the message, the according to the preset list that is set Do I Sound?” on pp 29. repeater is not opened. by the ATM switch set at each repeater 3K7NWS is a Seattle, Washington, repeater and site. Because the priority level can be KB7WUK is a Portland, Oregon, repeater on The Backbone Repeater Link designated in the header, voice signals the Evergreen Intertie. These examples as- One of the major advantages of the arrive in real time. This avoids the de- sume an identical system to the Evergreen D-STAR system is the full-duplex lays that happen with VoIP on the Intertie but based on D-STAR. !! 28 Sept/Oct 2003