A Transparent Repeater for Digital Communication Signals

Beiijamiii Friedlaiider Eli Pasteniak

Dept. Electrical Eiigiiieeriiig Bridgewave Commuiiicatioiis Inc.
Uiiiversity of California 3350 Thomas Road
Santa Cruz, CA 95064, USA Saiita Clara, CA

Abstract this as the Demod/Remod approach.

An interesting theoretical as well as practical q u e
A method is presented for re-transmitting a digi- tion is the following. Is there a way to convert the
tal communication signal over channels with differ- signal coming from the first channel into a form suit-
ent signal-to-noise ratios, without complete demodula- able for transmission on the second channel, without
tion and re-modulation (Demod/Remod). The method, demodulation and remodulation’? The Demod/Remod
which we call Signal Code Modulation (SCM), converts solution requires a relatively complex implementation,
samples of the incoming signal into digital and analog introduces signscant processing delay to the link, and
symbols whose characteristics are designed.to match the is specific to the modulation scheme being used. Thus,
channel. The SCM method trades bandwulth for SNR, it is of interest to look for a more transparent repeater,
in the classical manner, and achieves a performance which maintains the noise immunity of Demod/Remod.
close to that of conventional Demod/Remod methods. In this paper we describe a method which is able
However, it does so by perfonning relatively simple op- to accomplish this task efficiently, and achieve perfor-
erations on the sampled input signal, and does not re- mance close to that of Demod/Remod. The method
quire detailed knowledge of its fine structure (framing, involves converting samples of the input signal into a
coding, etc.). combmation of digital and analog symbols, which are
then transmitted through a conventional analog chan-
nel suitable for digital communications. The samples of
1 Introduction the input signal are then reconstructed from the analog
and digital symbols. The advantage of this technique
which we call Signal Code Modulation (SCM), is that
In some situations a digital communication signal
it is potentially simpler and cheaper to implement, and
needs t o be transmitted over several channels with dif-
is relatively transparent.
ferent bandwidth and signal-to-noise ratio (SNR) con-
ditions. Consider for example the case where the path
of the signal consists of two parts: in the first the 2 Signal Code Modulation
signal is transmitted over a cable, and in the second
over the air. The characteristics of t h e e two chan- To introduce the SCM technique we consider a spe-
nels are quite dBerent, requiring different modulation cific situation where it is desired to transmit a 2 5 6
techniques. For example, the high SNR cable environ- QAM signal over a channel suitable for a 16-QAM sig-
ment permits the use of high order modulation (say, nal.
256-QAM), while the lower SNR radio channel may re- Let z(n) be samples of a complex baseband 2 5 6
quire a lower order but higher bandwidth modulation &AM signal, sampled a t a rate of fs samples per sec-
(say, 16-QAM). The standard solution to this prob- ond. The 256-QAM signal has a symbol rate of f256
lem is t o demodulate and re-modulate the digital sig- symbols per second, and u s e a Square-Root Raised
nal as it goes from one channel to the other. In other Cosine (SRRC) filter with excess bandwidth factor
words, each part is treated as a separate digital com- a 2 5 6 , so that its effective (bandpass) bandwidth is
munication system which is optimized for that chan- +
B256 = (1 ~ 2 5 6f256. ) The baseband signal is con-
nel, and at the interface between channels the signal is tained in the frequency range [ - B 2 ~ ~ / B25&].
2, This
completely demodulated to reconstruct the original bit signal is passed through an anti-aliasing Elter with ef-
stream, which is then re-modulated. We will refer to +
fective bandwidth (1 P)B256/2 where p is related t o

0-7803-7576-9/02$17.00 0 2002 IEEE 1929

Authorized licensed use limited to: SRM University. Downloaded on November 24, 2009 at 07:45 from IEEE Xplore. Restrictions apply.
 the size of the passband-b-stopband transition region passed through a matched (SRRC) filter, sampled at
of the filter. Assuming sampling at the Xyquist rate, the symbol rate at the proper instants of time (assum-
wehavefs=(1+/3)B~5~=(1+P)(1+a256)f251i. I f @ ing perfect time synchronization), and passed through
and a256 are small, then fe is just slightly larger than a hard or soft decision device.
the baud rate. The value i.(n) can be transmitted similarly, as
Each sample ~ ( nwill ) be represented by a digital an “analog symbol”. Passing the sequence of values
1GQAM symbol, and an analog symbol representing through a transmit SRRC filter and sampling the re-
the difference between s(n)and the nearest 16.-QAM ceive SRRC filter at the right times, will reproduce the
constellation point. More precisely, let pi, i = 1, . . . , 16 same values, except for the addition of noise (again, as
denote the’valuesof a 18QAM constellation, where the suming for the moment perfect time synchronization).
real and imaginary values of the constellation points The digital and analog symbols can be multiplexed
are f4/2,&@S. Let 7 be a scale factor, to be de- in different ways for transmission on the given cban-
termined later, and let).(i denote the index which nel. Using tiisdivision multiplexing, the symbols are
minimizes).(.I - y q i [ . In other words, yqicnj is the interleaved, i.e, the sequence {qi(,,), gaIa(n)}is passed
constellation point nearest to the sample Let .I.(. through the SRRC filter and transmitted. The con-
s,(n) = z ( n ) - yqi(,,) be the difference between z(n) stant ga is a scale factor used to adjust the relative
and the nearest constellation point. Clearly, the index amplitudes of the digital and analog symbols. (An ad-
i(n)and the value r,,(n) provide an exact repwsenta- ditional scaling factor may be used t o multiply the in-
tion of ~ ( n-)see Figure 1. It is this information that terleaved symbols so as t o achieve the desired transmit
we wish to transmit over the channel. power. Without loss of generality we assume this factor
to equal unity.) Using frequency division multiplexing
the analog and digital symbols will be transmitted on
two separate channels.
At the receiver side the digital and analog symbols
are separated. The digital symbol is detected t o prm
duce an estimate I(.) of the index i ( n ) ,while the ana-
log symbol is divided by go t o produce an estimate
&(n) of the analog symbol s(n). Finally, the results
are combined to produce an estimate of the original
data, i.e., i ( n )= -yq;(,,] +
fa(.).
We will refer to the modulation technique described
above ils the SCM - AD method, where AD denotes
the fact that each input sample is converted into one
analog and one digital symbol. Later we will describe
other variations of the SCM system which use different
combinations of analog and digital symbols.
-On t Having described the SCM-AD method, a number of
questions arise. The key question is of course: what is
the performance of this technique and how does it com-
Figure 1. Decomposition of a measurement pare to Demod/Remod or, for that matter, to t r a n s
)
.
(
I into an analog value la(.) and a digital mitting the data z ( n ) directly over the channel? A s
suming that a convincing case can be made for using
value Pi(,,) (for 7 = 1).
this method, it is necessary to determine the optimal
values of the parameters 7 and go defined above. These
questions will be addressed in the next section as part
of the performance analysis of the SCM-AD method.
Note that i(n),which takes on one of 16 discrete
values, can be transmitted by using conventional 1&
QAM modulation. In the case of radio transmission, 3 Performance Analysis
for example, this involves the following steps: the digi-
tal symbols qi(,,) can be pasred through a SRRC filter, To understand the performance improvement of-
converted to analog, upconverted t o the desired RF fered by SCM, consider for the moment the case where
frequency and transmitted through the air. On the re- the digital symbols are transmitted through the chan-
ceiver side, the signal is down-converted to baseband, nel error free. Note that the analog symbol sa(.) prc-

1930

Authorized licensed use limited to: SRM University. Downloaded on November 24, 2009 at 07:45 from IEEE Xplore. Restrictions apply.
d u d by process described above, has a smaller vari- the SCM conversion parameter 7 , where the gain factor
ance than the original signal ~ ( n ) Consider . an unfil- is 2Olog,,(g,). We consider two situations. In the first
tered 256-QAM signal and the 16-$AM constellation WP asslime perfect carrier recovery so that the signal
used by the SCM method. The maximum value of constellation is k e d . In the second we assume that
za(n)in this case is four times smaller than the maxi- there is a frequency offset, so that the signal constella-
mum value of (.) Therefore, we can multiply ~ ( n ) tion is rotating at some unknown rate. This rotation
by g,, = 4, to give it the same variance as the origi- effects the variance of ~ ( nand) za(n).
nal signal ~ ( n )In. other words, if we compare direct We note that when there is no frequency offset the
transmission of 256QAM and of the analog symbol us- highest gain, 10.9dB, is obtained for y = 0.84. This
ing the same transmit power, we need to amplify the gain corresponds to ga = 3.47. When the signal has a
analog symbol by go = 4. At the receiver, the analog frequency offset, the highest gain is 10.4dB, obtained
symbol will have t o b e divided by ga to produce ?,(n), €or 7 = 0.86. In other words, there is only a small loss
and consequently, the measurement noise will also due to the frequency offset. This is important, because
divided by ga. Therefore, the reconstructed signal s(n) it means that we do not need to perform synchroniza-
will have a signal-twnoise ratio (SNR) which is g: times tion and carrier recovery on the incoming signal. We
better than the SNR experienced if r(n) were to be can sample it and transmit it ‘‘asis.”
transmitted directly over the channel. This figure was calculated for sampling a t the
Consider the case where the incoming signal r(n)are Nyquist rate and using 0 2 5 6 = 0.12. The results will
samples of an unfiltered 256QAM, and where y = 1 change depending on the sampling rate and the value
is the decomposition r,(n) = ~ ( n )7 ~ ( ~We - ) .as- of a2s6. However, in practice we want to use the low-
sume that go is.adjusted so that the resulting analog est sampling rate and a small excess bandwidth, to
signal has the same power as the original signal. This minimize the bandwidth of the SCM signal. Thus, the
is achieved by setting result shown here is a good representation for any prac-
tical SCM system.
The increased SNR comes at the cost of larger band-
width requirements. To see this, recall that the sig-
nal ~ ( nk) sampled at a rate of f8 = (1 P)&6 +
samples per second. For each sample we have an A
and a D symbol so the new symbol rate is Zf,. As-
suming an SRRC filter with excess bandwidth ascm,
the baseband SCM will have a bandwidth of B,,, =
+
2(1 %m)(1 P)BZ56. +
By comparison, when using DemodfRemod, each
256-QAM symbol is converted into two l 6 Q A M sym-
bols. Therefore, the required bandwidth will be B16 =
2(1 + Assuming that a 2 5 6 = a16,we have
&6 = B256 and therefore B,, = (l+,B)(l+aam)B16.
In other words, the bandwidth requirement of SCM-
AD is slightly larger than that of Demod/Remod, p r e
that B << 1 and aSm<< 1.
,I;‘.;.., ...... .. .. .. .. . .. . . ..... L... ... . ... . ... .
. . ’
.~
’’:
vided
Next we consider the performance of SCM for the
c s e where the digital symbols suffer transmission er-
rors. We will evaluate the Bit Error Rate (BER) and
Symbol Error Rate (SER) of the communication sys-
tem. Subscripts will be used to differentiate between
the different modulations considered here: BERIG,
Figure 2.The SCM gain factor 20 loglo(g,) as BERz56 and BER,, denote the BE% of K Q A M ,
function of the parameter 7,for a 256-QAM 256-QAM, and SCM, respectively. The SE% SER16,
input signal and a 16-QAM digital symbol, SERzs6 and SER,, are similarly defined.
with and without frequency offset.
As noted above, if the digital symbols are detected
perfectly, then SER,,(SNR) = sER~66(g: SNR). The
probability of error of a digital symbol is SER16(SNR).
Figure 2 depicts the SNR gain factor as fnnction of Note that an error in the digital symbol will results

1931

Authorized licensed use limited to: SRM University. Downloaded on November 24, 2009 at 07:45 from IEEE Xplore. Restrictions apply.
 in a large error in the reconstructed 256-QAM symbol, density NO. The SNR is deEned in the usual manner
and therefore the detected symbol will most likely be by
in error. It follows that
P
SNR= __ (9)
2N0B
SER,, -
(1 SER16(SNR))SER,S(g: SXR)
where B is the receiver bandwidth. The bandwidth
+ SERIG(SNR) (2) of the 256QAM input signal is BzS6, the bandwidth
of the 16-QAM demod/remod system is B16, and the
This is an approximate formula which represents an bandwidth of the SCM system is BSCM.We assume
upper bound on the symbol error rate. The approxi- that the demod/remod system operates at (approxi-
mation is due to the fact that we assume that a trans- mately) the same bandwidth as the input signal. How-
mission error will cause an error in the corresponding ever, the SCM system operates at a higher bandwidth
256QAM symbol with probability 1. In fact the p r o b because of the need to transmit both analog and digi-
ability of such an error may be less than one. tal sysmbols for each input symbol. Thus, for a given
The various symbol error rates are given by standard P and No the SNR of the SCM system will be smaller
formulas for &AM modulation with M constellation than that of the 256-QAM and 1 6 Q A M system. This
points 111, fact has to be kept in miid for proper interpretation of
the results presented next.
The BER formulas for all three systems are derived
SER,w(SNR) = 4 using the commonly used approximation that the bit-
error-rate is the symbol error rate divided by the num-
where
(3) ber of bits per symbol, log, M , and the 8 equals the
symbol SNR divided by log, M . For the 256-QAM
1 and SCM systems we have log,M = 6, and for the
Q ( I )= Z e r f c ( z / a ) (4) demod/remod system log, M = 4.

The corresponding bit error rate formula is

The formulas for the SER and BER as function of .b- ' d,
where Eb is the energy per bit, and No is the noise
spectral denisty, are

SER,(~)=~(I-&)
No
Q(&G) 3 log, bf Ea

(7)
and
Figure 3.Theoretical and Simulated Symbol
Error Rate for SCM-AD, 256-QAM input sig-
nal, 16-QAM digital symbol
(8)
The SNR in the equations-above needs to be care-
fuUy defined and understood. In order t o make a fair
comparison of different communication systems, we &s In Figure 3 we compare the SER of SCM computed
sume that they all operate using the same transmit using the formulas above with the SER obtained hy
power P , and experience the same noise power spectral a Monte Carlo simulation. As can be seen there is

1932

Authorized licensed use limited to: SRM University. Downloaded on November 24, 2009 at 07:45 from IEEE Xplore. Restrictions apply.
 communication system (e.g. SER < 0.01 for uncoded
communications).
In Figures 4 and 5 we depict the SER and BER
.:'r ,

. :.I. ,:. .: .:, . .\.,>...,... i,,:.. ............... ..:, ,

i
of 16-$AM (Demod/Remod), 256-QAM (direct t r a n s
mission) and SCM, computed using the theoretical for-
mulas above. The SCM performance is close to that of
the Demod/Remod method. It requires approximately
8dB less 3 than would be required if the same data
were transmitted directly. The results of Monte Carlo
. I simulations are very close to these theoretical formulas
' I and are, therefore, not shown.
. . . . . ..................... ,,:: ,,,I" ,,::" , .I: ..... " ' . ] To summarize the results so far: the SCM-AD
method makes it pmsible to transmit a 256QAM sig-
nal over a channel suitable for 16-QAM signals, at
.... :.. .... 1: ....... : ... ..... .:. ...........
I
the cost of approximately doubling the required band-
width. The performance and required bandwidth are
iD 13 28 21 w 3, *
SNRidq close to that of a Demod/Remod solution.
A more detailed analysis of the SCM system, includ-
Figure 4. Theoretical Symbol Error Rates for ing a study of the SNR gain factor for general input sig-
SCM-AD, 16-QAM DemodlRemodand 256- nals, the balancing of analog and digital gain factors,
QAM, for 256-QAM input and 16-OAMdigital and the impact of coding, is presented in [2].
symbol
4 Conclusions

The use of digital modulation techniques with dif-

ferent orders of modulation allows us to trade off
SNR for bandwidth. The traditional approach to
designing communication systems which operate over
... channels with different SNR characteristics is the De-
': i
i mcd/Remod method. The SCM technique presented
.. . . . . . . . . . .\ . < . ......... ;. ........ and analyzed here provides an interesting alternative
which mixes analog and digital symbols. Unlike the
Demod/Remod method, the SCM performs relatively
simple operations on the sampled input signal and does
not require detailed knowledge of its fine structure
. .
I (framing, coding, etc.). This increased transparency is
1 : :: .:.: .::.
I

. ::\>::.. ...... :..:: . ..: ... .:. ':':'.

2, . attractive in applications where the structure of the in-
?A! coming signal is not fully known, or when it may not be
i k e d over time. The greater flexibility of SCM comes at
I
............................. $ .......................... .....,.......... the ccst of reduced performance. However, the perfor-
, mance reduction is rather small, as was demonstrated
ZO 26
0" id? in this paper.

Figure 5. Theoretical Bit Error Rates for References

SCM-AD, 16-QAM DemdRemcd and 256-
QAM, for 256-QAM input and 16-QAM digital [l] John G . Proakis, Digital Communications, 3rd
symbol Ed., McGraw-Hill, 1995.
[Z)B. Friedlander and E. Pasternak, "A Transpar-
ent Repeater for Digital Communication Signals,"
submitted for publication.
a good match in the region where the symbol error
rates are sufficiently low to be useful for a practical

1933

Authorized licensed use limited to: SRM University. Downloaded on November 24, 2009 at 07:45 from IEEE Xplore. Restrictions apply.