IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS. VOL 9. NO. 6. AUGUST 1991
`
`909
`
`Performance Evaluation of a Multichannel
`Transceiver System for ADSL and
`VHDSL Services
`
`Peter S. Chow, Student Member, IEEE, Jerry C . Tu, Student Member, IEEE,
`and John M. Cioffi, Senior Member, IEEE
`
`study the performance of a multichannel mod-
`Abstract-We
`ulation method f o r two contemplated subscriber line d a t a ser-
`vices known a s asymmetric digital subscriber lines (ADSL) a n d
`very high-speed digital subscriber lines (VHDSL). I n the ADSL
`case, we find t h a t over all unloaded North American subscriber
`lines in o u r test set, a n unidirectional 1.536 M b / s d a t a r a t e
`service f r o m the end office t o t h e customer premises is possible
`on a single twisted pair a t a n e r r o r r a t e of lo-’ with at least a
`6 d B margin using coded multichannel modulation with suffi-
`cient transmit power. Furthermore, we find t h a t the proposed
`ADSL service can co-exist with basic-rate access’ ISDN (or
`voiceband analog services) on t h e s a m e twisted pair with o u r
`proposed system. I n the VHDSL case, d a t a rates in excess of
`100 M b / s can be transmitted reliably, a t a n e r r o r r a t e of lo-’,
`using uncoded multichannel modulation on a single twisted pair
`over a relatively short distance ( 5 150 feet) with a sufficiently
`high sampling r a t e ( ~ 2 4 MHz) a n d transmit power. I n this
`study, the dominant line impairments in the ADSL environ-
`ment include intersymbol interference (ISI), far-end crosstalk
`(FEXT) f r o m other ADSL services, spill-over near-end cross-
`talk (NEXT) f r o m baseband services if! the s a m e wire bundle,
`spill-over far-end baseband signal on the s a m e twisted pair due
`t o imperfect filtering, a n d additive white Gaussian noise
`(AWGN) f r o m such sources as electronic a n d thermal noises.
`In the VHDSL environment, ISI, F E X T a n d NEXT f r o m other
`VHDSL services in the s a m e wire bundle, as well a s AWGN,
`a r e included. Finally, we show t h a t a cost-effective multichan-
`nel transceiver design that has been suggested f o r high-speed
`digital subscriber lines (HDSL) service will also work well for
`the proposed ADSL a n d VHDSL services with only minimal
`modifications.
`
`I. INTRODUCTION
`ITH the advent of high-speed digital subscriber lines
`(HDSL) technology in recent years, a number of
`new high-speed transport concepts have been proposed in
`the communications industry. Among these newly pro-
`posed transport concepts, asymmetric digital subscriber
`lines (ADSL) and very high-speed digital subscriber lines
`(VHDSL) are two of the most promising future data trans-
`
`Manuscript received November 9. 1990; revised May 15, 1991. The
`work of P. S. Chow was supported in part by a National Science Founda-
`tion Graduate Fellowship. The work of J . M . Ciofi was supported in part
`by Bell Communications Research and the University Technology Transfer
`Institute. Part of this paper was presented at the IEEE HDSL Workshop
`’91, Sunnyvale, CA, June 19-20, 1991.
`The authors are with the Information Systems Laboratory, Department
`of Electrical Engineering, Stanford University. Stanford. CA 94305.
`IEEE Log Number 9101635.
`
`mission services in today’s market. In this paper, our em-
`phasis is on the description and performance evaluation
`of one type of HDSL transceiver technology that can be
`used to achieve reliable data transmission for both ADSL
`and VHDSL services.
`The proposed ADSL service will support a 1.536 Mb/s
`data rate on standard twisted-pair telephone lines, unidi-
`rectionally, from the Central Office to the customer prem-
`ises. The term “asymmetric” in ADSL refers to a high
`data rate in one direction only, as ADSL is distinguished
`from HDSL, the latter of which is bidirectional and will
`only be offered for the restricted set of loops within the
`so-called carrier serving area (CSA)’. Furthermore, HDSL
`is intended for conventional T1 or,DSl data rate services,
`while ADSL is a consumer service with the intended ap-
`plication being the transmission of compressed TV-quality
`video (see [2]) with distribution over almost the entire loop
`plant, including those outside of the CSA. Lastly, ADSL
`is also currently being considered, for economiCal rea-
`sons, to be superimposed on the same single twisted pair
`that delivers basic-rate access ISDN service [or possibly
`the plain-old telephone service (POTS)] while HDSL will
`be a substitute service for basic-rate access ISDN or
`POTS.
`Because of the asymmetry in the transmission system
`for ADSL, near-end crosstalk (NEXT) from one ADSL
`service to another cannot occur, and we will show that in
`the ADSL environment, the dominant line impairments
`interference (ISI), far-end crosstalk
`are intersymbol
`(FEXT) from other ADSL services, spill-over near-end
`crosstalk (NEXT) from baseband services, spill-over far-
`end baseband signal on the same twisted pair, and addi-
`tive white Gaussian noise (AWGN) from such sources as
`electronic and thermal noises. The absence of near-end
`crosstalk due to other ADSL sefvices in the same wire
`bundle significantly improves the data transport capability
`of the twisted pair and allows 1.536 Mb/s transmission
`on all loops in our test set with adequate margin when a
`sufficiently powerful transceiver is used.
`VHDSL, on the other hand, is a high-speed data trans-
`port concept that serves as a component of the eventual
`
`‘The CSA consists mainly of loops that are less than two miles in length
`(see Bellcore RF1 90-03 [ 11).
`
`0733-871619110800-0909$01 .OO W 1991 IEEE
`
`

`
`910
`
`IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 9. NO. 6. AUGUST 1991
`
`migration to fiber within the loop plant. VHDSL consid-
`ers full-duplex transmission of data at rates significantly
`higher than the HDSL rate but over only that segment of
`the loop plant located between the pedestal and the cus-
`tomer premises, and it presumes a high-speed media (such
`as fiber) from the end office to the pedestal. Data rates of
`interest include those sufficiently high to sustain computer
`network applications, such as 8, 10, and 16 Mb/s, the
`T 3 rate of 44.736 Mb/s, the OC-1 rate of 51.84 Mb/s,
`the FDDI rate of 100 Mb/s, and the OC-3 rate of 155.52
`Mb/s.
`Typical VHDSL channels are short drops ( 5 150 feet)
`of unshielded twisted pair (26 gauge or better) from the
`pedestal to the customer premises. There are usually two
`such pairs into the average customer premises, and these
`two pairs can crosstalk into one another if both are used.
`IS1 on such channels is not as severe as those encountered
`in HDSL or ADSL but is still significant at the data rates
`of interest, and we find near-end crosstalk (NEXT) to be
`more limiting than far-end crosstalk (FEXT) when both
`pairs are used.
`In this paper, we focus on multichannel modulation and
`show that it is an excellent method for delivering reliable
`high data rates to the customer, both in terms of perfor-
`mance and cost, for ADSL and VHDSL. In Section 11, we
`describe the loop plant characteristics for both the ADSL
`and the VHDSL environments. We also describe the pro-
`visions made for possible co-existence of ADSL with
`other analog and digital baseband data services and the
`inclusion of a reverse channel capability in ADSL. In
`Section 111, we briefly review the characteristics of a mul-
`tichannel modulation system that is intended for use with
`ADSL as well as VHDSL. In particular, we use discrete
`Fourier transform in the transceiver design with the goal
`of maximizing performance and minimizing computa-
`tional requirements. In Section IV, we investigate achiev-
`able data rates both within and outside of the CSA for
`ADSL and find that all loops in our test set will reliably
`support ADSL service with at least a 6 dB margin using
`our specific version of coded multichannel transceiver. In
`Section V, we investigate achievable data rates for
`VHDSL with varying system parameters, and we find that
`with a very high signaling rate (24 MHz) and sufficient
`transmit power, we can reliably transmit over 100 Mb/s
`through VHDSL lines even without an additional trellis
`code. Finally, we summarize our findings in Section VI.
`
`11. LOOPS AND SIGNAL CHARACTERISTICS
`A. ADSL Transmission Characteristics
`
`The proposed ADSL service is closely related to HDSL;
`however, unlike HDSL, ADSL provides unidirectional
`high data rate service to consumers within the carrier
`serving area (CSA) and possibly outside of the CSA. At
`the present time, there are no established design rules for
`ADSL, though several key characteristics of the ADSL
`environment have been identified as follows [3]:
`
`1) All loops are nonloaded.
`2) All loops consist of 26 gauge or coarser cables,
`either used alone or in combination with other gauge ca-
`bles.
`3) The maximum allowable loop length, including
`bridged taps, is 18 kft.
`4) A reverse channel with a maximum data rate of 9.6
`kb/s is to be provided for control purposes.
`In addition, it would be economically advantageous if
`the ADSL service can be superimposed on the same line
`that delivers basic-rate access ISDN or analog baseband
`services, e.g., POTS. For this reason, though it is not a
`formal design rule for ADSL, we will evaluate a multi-
`carrier system that delivers the unidirectional 1.536 Mb/s
`data rate without occupying or interfering the lower 50
`kHz of the frequency spectrum. Furthermore, an impor-
`tant design objective for the authors is that the existing
`HDSL transceivers should be easily modified to handle
`ADSL service, e.g., via simple software changes, so that
`minimal design impact is to be incurred while imple-
`menting ADSL as a subset of HDSL.
`
`B. Representative Loops in the ADSL Environment
`The set of loops under study is shown in Fig. 1. They
`are representative of “lossy” loops within and outside of
`the CSA [4], [5]. The impulse response and power spec-
`tral density’ characteristics for these channels have been
`determined with data in [6], using a modified version of
`the LlNEMOD program3. A pole-zero model is added to
`the response of each loop to eliminate the dc component
`and to simulate the effects of the transformer coupling that
`exists at both ends of the twisted pair. This pole-zero
`model consists of a double-zero at dc and a double-pole
`that makes the power gain of the transformer equals to
`- 6 dB at 300 Hz.
`We will find, as a general result, that channels outside
`of the CSA perform significantly worse than those within
`the CSA. This is to be expected as channels outside of the
`CSA suffer much larger signal attenuation because they
`are longer in length. Cable attenuation characteristics of
`various channels in the frequency domain are illustrated
`in Fig. 2, which compares the power spectral densities of
`loops outside of CSA to one of the “worst-case” CSA
`channels (channel 6 ) . We should point out that these
`power spectral densities are calculated based on source-
`to-load loss. Therefore, power spectral densities of the
`same loops calculated based on insertion loss should be
`scaled 6 dB higher uniformly across the entire frequency
`band, assuming that the source and load impedances are
`matched. We will use these power spectral densities based
`on source-to-load loss in our computer analysis for
`achievable throughput of our proposed system, which im-
`plies that we have tacitly included a 6 dB noise margin
`
`’We define the “power spectral density” of a channel as the magnitude
`squared of the Fourier transform of the channel impulse response.
`’We thank Prof. D. G . Messerschmitt of the University of California at
`Berkeley for making the source code of this program available to us.
`
`

`
`CHOW ef al.: PERFORMANCE EVALUATION OF A MULTICHANNEL TRANSCEIVER SYSTEM
`
`91 I
`
`( U D
`
`IXQVn.6
`
`IWn.6
`
`-8
`
`Mrm
`lDmm
`IXUn.6
`Fig. 1 . ADSL loops under study within and outside of the CSA.
`
`PSDs of Channels Inside and Outside of CSA (6 dB Margin Included)
`
`- - - - CSAChannel6
`- Outside CSA Channel E
`- - - Outside CSA Channel D
`Outside CSA Channel C
`
`C. ADSL Line Impairments
`Empirical studies indicate that crosstalk is likely to be
`the limiting impairment on two-way transmission at the
`ISDN frequencies of interest. It has been shown that the
`crosstalk phenomenon can be accurately modeled using
`only two terms, namely the near-end crosstalk (NEXT)
`and the far-end crosstalk (FEXT) [8]. The NEXT and
`FEXT terms are identified in Fig. 3. In the ADSL envi-
`ronment, data is transmitted unidirectionally . Therefore
`there will be no NEXT term in the frequencies of interest
`due to other ADSL services in the same wire bundle. (We
`have also assumed that there will be no T 1 or HDSL sys-
`tems in operation within the same wire bundle, for they
`would be extremely damaging to an ADSL system due to
`their large high frequency content.) However, there will
`be a significant amount of spill-over NEXT from bidirec-
`tional baseband services in the same wire bundle due to
`nonideal filtering. This NEXT term can be modeled with
`a coupling function of the form [SI:
`
`(1)
`l H N E X T ( f ) 1 2 = K N E X T f 3’2
`where f is frequency in Hz and K N E X T is determined
`through empirical measurement. In our study, we will as-
`sume the worst-case scenario of 49 crosstalkers all due to
`[9],
`basic-rate access ISDN service, where K N E X T =
`[lo]. The input power spectrum to this coupling function
`is the transmitted spectrum of the baseband service. For
`basic-rate access ISDN service over AT&T DSL’s, square
`pulses are passed through a second-order lowpass filter
`with the 3 dB point at 40 kHz and at least a 50 dB/decade
`rolloff after 50 kHz [ 1 11. The actual signal voltages used
`are f 2 . 5 volts and 4-2 volt, leading to an average power
`of approximately 25 mW with a 135 Q load impedance.
`We will use this transmit spectrum and power level for
`BA-ISDN in our computer evaluation, though this is not
`the worst-case transmit spectrum. In particular, if sinc
`pulses are used instead of square pulses, we would have
`a flat instead of sinc squared spectrum.
`In addition to spill-over NEXT, FEXT also exists in
`the ADSL environment, and it can be modeled with a cou-
`pling transfer function of the form [12]:
`
`I H F E X T ( ~ , f ) I = K F E X T ~ I C( f ) I ’f ’
`(2)
`where I C( f ) l2 is the channel power spectral density func-
`tion as plotted in Fig. 2, d is the length of the cable in
`kft,fis frequency in Hz, and K F E X T is determined through
`empirical measurement. In our study, we will mostly as-
`sume that K F E X T =
`[13]-[15], though we will also
`investigate the effect of varying the level of K F E X T at sat-
`uration power level in Section IV-B.
`In addition to crosstalk noise, there are several other
`impairments that will degrade the performance of any sys-
`tem operating over twisted copper pairs at ADSL rates.
`Because we have assumed that the ADSL service is su-
`perimposed on the same wire that delivers baseband BA-
`ISDN, the spill-over far-end signal from BA-ISDN on the
`same wire will be received along with the ADSL signal,
`
`-loot
`-1201
`
`0.5
`
`1
`
`1.5
`2
`Frequency (Hz)
`X I 0 5
`Fig. 2. Power spectral densities of representative ADSL loops under study.
`
`I
`
`2.5
`
`3
`
`(as is now required in HDSL applications) in any envi-
`ronment where the dominant source of noise does not de-
`pend on the channel characteristics. This is the case for
`HDSL, where the channel-independent near-end crosstalk
`is the dominant source of noise [7], and we will show in
`Section V-B that this is also true in the VHDSL environ-
`ment. In the case of ADSL, however, the dominant source
`of impairment in most of our test loops turns out to be the
`channel-dependent far-end crosstalk, which is propor-
`tional to the channel transfer function. Thus, the factor of
`6 dB margin does not apply. We will describe the various
`channel impairments for both ADSL and VHDSL in more
`detail in Sections 11-C and 11-F, respectively, and we will
`evaluate the consequences of including a 6 dB noise mar-
`gin explicitly on the ADSL system in Section IV-B.
`
`

`
`912
`
`Subscriber #1
`
`Subscriter #2
`
`Subscriber #N
`
`Fig. 3 . Near-end and far-end crosstalk
`
`and this spill-over “signal” must be treated as noise by
`the ADSL receiver. Electronic noise, including quanti-
`zation noise from the A/D converter, and thermal noise
`in the analog portion of the receiver can be modeled as
`additive white Gaussian noise (AWGN). We will assume
`a fixed level of AWGN at -140 dBm/Hz in our study,
`which represents a fairly conservative estimate [5]. In-
`ductive noise at 60 Hz and its harmonics generally do not
`cause a problem at ADSL rates [l]; therefore, it will be
`ignored in our study. Residual echo noise is also negli-
`gible since there will be no echo due to ADSL at the re-
`ceiver end as data is transmitted unidirectionally, and re-
`sidual echo from baseband service is assumed to be
`sufficiently cancelled and filtered out. Impulse noise
`caused by switching transients. lightning, and electrical
`machinery is less well understood, and usually an opera-
`tional margin of 6-12 dB is placed on a system to handle
`such infrequent but high-peak noise. We shall not include
`impulse noise in our model. However, the block process-
`ing nature of our multicarrier system is advantageous
`when dealing with this type of short but intense noise,
`since the noise energy contained in the impulse is effec-
`tively spread over the entire block and the blocklength
`used is typically in the order of several hundred times that
`of the duration of the impulse (see [7]). Impulse noise of
`long duration, on the other hand, is hopefully eliminated
`by the high-pass filter in ADSL. Lastly, intersymbol in-
`terference (ISI) is inherent in ADSL loops since the trans-
`fer characteristics of the channels are nonideal. We shall
`describe how IS1 can be mitigated using a multicarrier ap-
`proach in Section 111.
`
`D. VHDSL Transmission Characteristics
`The proposed VHDSL service is an enhancement [16]
`of the presently developing HDSL service. VHDSL pro-
`vides reliable, bidirectional data transmission at rates of
`10 Mb/s or higher over only relatively short distances.
`Loops intended for use with VHDSL service are those
`that are located between the pedestal and the customer
`premises. These are generally no more than 150 feet in
`length and no smaller than 26 gauge in size. As in the
`
`IEEE JOURNAL O N SELECTED AREAS IN COMMUNICATIONS, VOL. 9. NO. 6. AUGUST 1991
`
`case of ADSL, VHDSL is a new transport concept with
`no established design rules yet. Some of the target data
`rates that may be desirable for VHDSL applications in-
`clude:
`Current Ethernet standard rate = 10 Mb/s
`DS-3 (digital signal, level 3) = T 3 rate = 44.736
`Mb/s
`OC-1 (optical carrier, level 1) = 51.84 Mb/s
`FDDI rate = 100 Mb/s
`OC-3 (optical carrier, level 3) = 155.52 Mb/s
`
`E. Representative Loop in the VHDSL Environmenr
`In this study, we only consider the worst-case VHDSL
`loop, i.e., 150 feet of 26 gauge wire. Furthermore, we
`assume that PIC cables operating at 70°F with matched
`source and load resistances of 110 Q are used and that
`within this short 150 feet line there are no bridged taps or
`wire gauge changes. As in ADSL loops, a transformer is
`added to both ends of the cable to eliminate the dc com-
`ponent in the frequency response. Again, the effects of
`the transformer coupling is simulated by a pole-zero model
`that consists of a double-zero at dc and a double-pole that
`makes the power gain of the tranformer equal to -6 dB
`at 300 Hz. Using the modified version of LINEMOD with
`data in [6], we can determine the impulse response and
`power spectral density characteristics for this channel. We
`found that the maximum useful bandwidth of this worst-
`case loop is around 12 MHz (see Fig. 4). Thus, a maxi-
`mum signaling rate of 24 MHz will be used in this study.
`
`F. VHDSL Line Impairments
`As in the case of ADSL, NEXT, and FEXT are two of
`the most serious line impairments encountered in VHDSL.
`The coupling function for NEXT is given by (1) in Sec-
`tion 11-C. In [9], Lin estimated that K N E X T =
`for
`the 49-crosstalker case from data in a Bellcore Technical
`Reference [lo]. In the case of VHDSL, however, the
`twisted pairs are most likely unbundled. Therefore, in-
`stead of 49-crosstalkers, there will usually be a maximum
`of only one crosstalker, as existing customer lines often
`contain two twisted pairs into each customer premises.
`We will assume that K N E X T = $j x lo-” = 2 X
`for our test loop. Note that if we assume there are nor-
`mally only 6 dominant crosstalkers in a 50-pair bundle,
`then the VHDSL K N E X T with one crosstalker should be
`closer to X
`We will investigate the effect of vary-
`ing the level of K N E X T in Section V-B. The coupling func-
`tion for FEXT is given by (2) in Section 11-C. As in the
`case of K N E X T , K F E X T is determined through empirical
`measurement [ 131-[ 151. In this study, we will assume that
`KFEXT x d = & x 0.15 x
`= 3 x
`for the
`test loop, since there will only be one far-end crosstalker
`instead of 49 and d = 150 ft = 0.15 kft. The performance
`of DMT for K F E X T varying over several orders of magni-
`tude are discussed in Section V-B. Besides NEXT and
`FEXT, the transmitted data may also be corrupted by in-
`teraction crosstalk and apparatus crosstalk at VHDSL fre-
`
`I
`
`--
`
`..
`
`

`
`CHOW cf a l . : PERFORMANCE EVALUATION OF A MULTICHANNEL TRANSCEIVER SYSTEM
`
`913
`
`Power Spectral Density of IS0 Feet of 26 Gauge Wire Loop
`
`-1401
`
`A. The Discrete Multitone System
`The fundamental concept of multicarrier modulation is
`the conversion of a data transmission channel with inter-
`symbol interference (ISI), and possibly crosstalk and/or
`colored noise, into a set of parallel, independent, and
`ISI-free subchannels. In [20], Bingham gives a compre-
`hensive tutorial on the various multicarrier modulation
`methods, and the specific modulation technique evaluated
`in this study is known as the discrete multitone (DMT)
`modulation. DMT modulation is based on frequency-di-
`vision partitioning of the channel spectrum using the dis-
`crete Fourier transform, and it is an enhanced version of
`what appears in [21]. We shall not attempt a detailed de-
`scription of the DMT modulation in this section, but rather
`we refer readers to the companion paper [7] by J . S. Chow
`and two of the co-authors of this paper on the performance
`evaluation of the DMT system for high-speed digital sub-
`scriber lines (HDSL) in this same issue.
`
`B. Trellis Code Concatenation
`One attractive feature of multicarrier modulation is that
`since the independent subchannels are memoryless, coset
`codes (trellis codes) can be concatenated to the modula-
`tion structure, and the coding gain of these powerful codes
`can be realized by the system. In [22] and [23], Fomey
`characterized most of the known good codes for bandlim-
`ited channels as coset codes. In the DMT system studied
`here, we will use a method of code concatenation de-
`scribed in [24] known as “coding down the block,” in an
`effort to achieve a good tradeoff between hardware com-
`plexity and decoding latency. This method was also in-
`herent in [25] and in a recently filed patent application by
`Decker of Telebit [26]. In this method, a single encoder/
`decoder pair is used and several coded symbols from the
`encoder are concatenated together to form the multidi-
`mensional transmit vector that is to be processed by the
`multicarrier modulator. Each orthogonal dimension of the
`transmit vector is assigned to succeeding subchannels of
`the multicarrier system until all subchannels are used. The
`details of the coset code concatenation procedure are given
`in [24].
`
`C. Channel Identijication and Protocol
`An accurate estimate of the channel response is neces-
`sary in discrete multitone modulation, and this informa-
`tion must be available to both the transmitter and the re-
`ceiver in order for the bit allocation algorithm and the
`pole-cancelling filter to function properly. The channel
`identification process takes place during the initial startup
`procedure. We will now briefly summarize the general
`protocol used by the DMT for ADSL and VHDSL ser-
`vices.
`After receiving a request for transmission signal from
`one subscriber, predefined test patterns are transmitted to
`verify that the channel is indeed operational. Timing and
`synchronization are established at this time. Then a pre-
`defined pseudorandom training sequence is sent from the
`
`-200 ‘
`
`0 5
`
`\
`I S
`1
`Frequency (Hz)
`x10’
`Fig. 4. Power spectral density for 150 feet of 26 gauge copper twisted pair
`loop.
`
`2
`
`I
`2 5
`
`quencies. However, the specific effects of interaction
`crosstalk, which is the coupling between two pairs in-
`volving other pairs, and apparatus crosstalk, which takes
`place at terminals, splices, and cross-connects, are not
`well understood [ 171. Therefore, they will not be included
`here.
`In addition to various forms of crosstalk, there are sev-
`eral other potential impairments that will degrade the per-
`formance of a VHDSL system. Residual echo noise does
`exist in VHDSL loops, and it is typically modeled as ad-
`ditive white Gaussian noise (AWGN). Assuming an input
`signal power of around 20 mW, a typical VHDSL loop
`will attenuate the signal by approximately 10 dB, result-
`ing in a total received signal power of 3 dBm. With an
`excellent echo canceller that can reduce the echo to a level
`of 40 dB below the received signal, we expect a total noise
`power of around -37 dBm across a two-sided bandwidth
`of 24 MHz, yielding a noise power spectral density of
`approximately - 110 dBm/Hz. Therefore, we will fix the
`level of AWGN due to residual echo at a constant level
`of - 110 dBm/Hz throughout this study. Other sources
`of AWGN noise, such as electronic noise and thermal
`noise, are less damaging. We will assume a fixed level of
`electronic noise at - 140 dBm/Hz in our study [ 5 ] , which
`represents a fairly conservative estimate and can be ig-
`nored since it is much lower than the AWGN level of the
`residual echo noise. The effects of inductive noise and
`impulse noise in VHDSL are similar to those in ADSL,
`and lastly, IS1 can be mitigated by using a multicarrier
`modulation approach described in Section 111.
`
`111. MULTICHANNEL MODULATION
`
`Recently, various multicarrier systems [ 181-[21] have
`been proposed to transmit data reliably in the presence of
`severe intersymbol interference (ISI) for digital sub-
`scriber line applications. In this study, we focus on a spe-
`cific implementation of multicarrier modulation, known
`as the discrete multitone (DMT) modulation [2 11.
`
`

`
`914
`
`IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS. VOL. 9. NO. 6, AUGUST 1991
`
`Bit Allocation for CSA Channel 6 (12 dB Noise Margin)
`
`14------7
`
`l2 t
`
`:.. .
`-._
`, --. ,
`
`
`-
`Uncoded System
`- - ~ -
`5 dB Trellis Coded System
`
`2t
`
`i
`
`transmitter for channel identification. An ARMA (auto-
`regressive/moving average, or pole-zero) model for the
`channel is derived from the received training sequence via
`a least squares fit to minimize the prediction error be-
`tween the ARMA model and the received channel out-
`puts. The receiver can now set the coefficients for the pole-
`cancelling filter, compute the SNR for each frequency-
`indexed subchannel, and perform the bit allocation algo-
`rithm. The bit assignments are then sent back to the trans-
`mitter via a reliable feedback channel, which can be one
`of the subchannels operating in the reverse direction using
`highly redundant repetition code to ensure reliability.
`After frame synchronization is established, the DMT
`transceiver system is ready for user data transmission. In
`the ADSL environment, it should be stated that the infor-
`mation sent from the customer to the Central Office during
`this startup period will not be in the lower 50 kHz of the
`frequency band; thus, it will not interfere with existing
`baseband services on the loop. This requires the use of at
`least one subchannel in the opposite direction, further-
`more, this subchannel is silent in the forward direction.
`For our proposed system, we will use two of our fre-
`quency bins, occupying the frequency spectrum from 50
`to 52.5 kHz, for this purpose, which will also serve as a
`reverse channel for system control and monitoring func-
`tions during normal operation.
`
`IV. PERFORMANCE EVALUATION
`FOR ADSL
`In this section, we analyze the performance of the DMT
`transceiver for ADSL. The method of computer evalua-
`tion is given in [7], and we shall not repeat it here.
`
`A . System Parameters
`The system that we evaluate has a blocklength of 512
`and a signaling rate of 640 kHz. These system parameters
`are only representative and can be changed easily. In par-
`ticular, we note that a longer blocklength (say 1024) or a
`higher signaling rate (for example, 800 kHz) will yield
`somewhat higher data rates at the cost of longer decoding
`delay and higher complexity. In [27], Aslanis and Cioffi
`showed the capacity of loops inside the CSA, subject to
`a chosen signaling rate, will continue to increase until the
`signaling rate approaches 600-800 kHz, and we based our
`choice of signaling rate on this result. However, that study
`included NEXT from HDSL, whereas in ADSL we need
`not consider this impairment. Thus, higher sampling rates
`could yield yet better performance. Our choice of block-
`length, on the other hand, reflects the current technology
`in terms of feasible computational complexity and cost,
`which will be discussed briefly in Section IV-C.
`
`B. Performance Results
`The performance of our proposed DMT transceiver in
`the ADSL environment is evaluated as a function of trans-
`mit power. Using the DMT’s frequency-specific bit allo-
`cation algorithm, we can shape the transmitted spectrum
`
`0 ’
`0
`
`L
`0.5
`
`1
`
`2
`1.5
`Frequency (Hz)
`xi05
`Fig. 5 . Uncoded and coded bit spectral efficiency for CSA channel 6 with
`20 mW of power and 12 dB noise margin.
`
`2.5
`
`3
`
`1
`3.5
`
`for the ADSL environment. We show in Fig. 5 an ex-
`ample of bit allocation for CSA channel 6 at the maximum
`achievable data rate ( > 1.536 Mb/s) using 20 mW of
`transmit power with BER =
`and explicitly allowing
`a 12 dB noise margin. We note that the lower 50 kHz of
`the frequency band has been left free of transmission for
`spectral compatibility with existing baseband services,
`such as basic-rate access ISDN or voiceband POTS. Also
`note that the actual numbers of bits assigned to frequency
`bins between 50 and 75 kHz are smaller than those for
`bins immediately after 75 kHz, since spill-over noise from
`baseband services is extremely damaging in this fre-
`quency range.
`We evaluated the system for transmit power level rang-
`ing from 1 to 200 mW (0-23 dBm) for both CSA loops
`and loops outside of CSA. The achievable data rates at a
`bit error rate (BER) of lo-’ are shown in Figs. 6 and 7.
`We next investigated the performance of the DMT system
`with a concatenated trellis code. We assumed the coding
`gain achievable using an 8-dimensional, 64-state trellis
`code, i.e., approximately 5 dB of coding gain. The re-
`sulting data rates are shown in Figs. 8 and 9. Since the
`exact transmission impairments over digital subscriber
`lines are not completely understood at this time, an op-
`erational margin of 6 dB or more is typically required for
`a system to handle any unexpected source of noise. In
`Figs. 10 and 11, we have plotted the achievable data rates
`of a coded DMT system with a 6 dB operational margin.
`Our evaluation shows that all channels in our test set can
`support 1.536 Mb/s data rate transmission in the ADSL
`environment using a DMT transceiver with reasonable
`transmit power. In fact, using 10 mW of power, we can
`transmit over 3.2 Mb/s over the worst CSA channels
`using a coded DMT system (Fig. 8). The performance
`level can be further improved if a steeper lowpass filter is
`used in the BA-ISDN service, say a lowpass filter with a
`the assumed 50
`70 dB/decade
`rolloff
`instead of
`dB/decade rolloff, to isolate the baseband signal from the
`passband ADSL service.
`
`

`
`CHOW et al.: PERFORMANCE EVALUATION OF A MULTICHANNEL TRANSCEIVER SYSTEM
`
`X I 0 6
`3.2,
`
`Uncoded Inside CSA (No Margin)
`
`z;,, . .
`,.,
`
`,,/
`,.i”
`
`0 e.
`
`D
`
`p
`
`Y
`D
`
`- CSA Channel 1
`CSA Channel 2
`....
`CSA Channel 3
`~ . . CSA Channel 4
`-
`CSA Channel 5
`CSA C