Document Text (Pages 31-40) Back to Document

Adaptive secure data transmission method for OSI level 1

by Lallo, Pauli, PhD


Page 31

Fig. 3.3 Measured attenuation distortion of ADM-channel [Lal97a]

Measurements were made during 1990-1992 in The Signal School in Finland to find out
some basic characteristics of the adaptive delta modulated speech channel of the 16 kbps
network. The results are shown in Figure 3.3 [Mdd92]. The main channel and data transmission
measurements were continued in the Signal Schools, Finland from 1991-1994 with a
HP3567A measuring instrument made by the Hewlett-Packard Company, a data communication
analyzer 2871 of Marconi and a Datatest 3 of Navtel Canada Inc. The HP3567A instrument
has a built-in Fast Fourier Transform (FFT) capability. The analyzer 2871 and the
Datatest 3 perform a bit error rate analysis.

Bandwidth for Voice and Data Transmission

The 16 kbps ADM-channel voice bandwidth is about 2600 Hz and is thus not equivalent to
the ITU-T requirements [Itu88]. The equivalent measured bandwidth of the 16 kbps and 32
kbps ADM systems is presented in Figure 3.3. The 32 kbps bit rate makes the ADM-channel
more compatible with the ITU-T attenuation distortion requirements than 16 kbps.

The measurement results in Figure 3.4 show that the attenuation distortion of a 16 kbps
ADM-channel is not in the limits of ITU-T requirements for analog data transmission. The
result was expected based on the attenuation distortion requirements of the ADM-channel
defined in the reference [Eur86].

31


Page 32

Fig. 3.4 Measured attenuation distortion versus ITU-T M 1020 requirements
[Lal97a]

Message Intelligibility in DM and PCM

For voice communications, the message intelligibility is the performance measure of greatest
interest to the users [Sha79]. Then the performance of DM is better than the performance of
PCM while human communication is possible with lower S/N-ratio and bit error rate.
Experiments have shown that an error probability Pe of the order 10-1 does not affect the
intelligibility of voice signals in DM, whereas Pe as low as 10
-4 can cause serious errors
leading to threshold in PCM [Sha79]. In PCM the weight of the detection error depends on
the digit location 27
, 26
, …20
. Errors in the detection of the first digits have a greater effect
on the output signal waveform than errors in the later digits.

It was found in the analysis made at the Signals School in 1990-1992 that conversation in
delta modulated channels was possible to the point where the channel synchronization was
lost at bit error rate >10
-3 at S/N=9 dB.

In data communications the phase detection is difficult in delta modulation systems because
of many reasons:
- Symbol timing in asynchronous systems.
- Slope overload in delta modulation.
- Granular distortion.

ADM versus PCM

Comparing adaptive delta modulation (ADM) with pulse code modulation (PCM) one could
find that the latter also follows the phase of a signal in a better way. This is based on the 8-
bit code of the PCM, which adapts faster to the input signal than a one-bit code of ADM. In
PCM one knows quite well the zero points of the signal which correspond to the phase = 0.
In ADM the zero points are not as well defined because of the slope overload, continuous
variable step size and the operative delay of the leaking integrator used in the delta coder.

32


Page 33

3.3. Measurement Results and Analysis

The measurements of data transmission were made over granular noise channels and later in
field tests multi-path channels as reported in papers [Lal97a], [Lal97b], and [Lal02b]. Several
problems were found in analog data transmission in granular networks based on 16 kbps delta
modulation (DM) voice coding:
- Modern ITU-T high-speed modems or telefax do not work there.
- Even low speed (1200-2400 bit/s) FSK and PSK ITU-T modems (V.23 and V.26) have high

bit error rates.
- Delay times grow too high in packet switching.

The measurement results include
1) Magnitude and phase response of the granular channel.
2) Distortion characteristics of the ADM-channel.
3) Data transmission quality measurements using standard modem waveforms.
4) Analysis of block error probability in packet switching.

Figure 3.3 presents the measured attenuation distortion of a digital channel (granular ADMchannel).
Attenuation depends heavily on digit rates. The attenuation at a digit rate of 16 kbps is
not in the frames of the ITU-T recommendations for data transmission. Data transmission
measurements over these channels are described in the paper [Lal97a]. Table 3.1 presents the
considered standard modem types, which were tested. V.23 and V.26 modems were the only
acceptable models. The bold spectrums in Table 1.1.correspond to the measured 16 kbps
granular noise channel bandwidth shown in the Figure 3.3.

Fig. 3.5 Measured bit error rate of V.26 modem data transmission
A 24-hour test

However, the data transmission test shows poor results, Figure 3.5. Modeling and simulation
methods are used to find better analysis and solutions in the next chapter. Figure 3.6 presents
the result using the ADM (adaptive delta modulation) as a granular channel. It is found the

33


Page 34

interesting capability of ADM to maintain the bit error rate level constant. The reason was
expected and obvious due to the adaptation of the step size according to the signal level. The
bit error (BER) tests of two modem types are presented in Figure 3.6. BER is in the range 10-
4
…10

-3

. The recommendation of the digital 16 kbps networks [Eur86] describes only digital
data transmission classes, which include error correction methods and faster bit rates (9.6 –
16 kbps).

Fig. 3.6 Measured bit error rates of data transmission [Lal97a]

34


Page 35

Packet Switching Problems in Internet

Figure 3.7 illustrates BLER (block error rate) results with FSK waveform at low S/N ratios,
paper [Lal97a]. The retransmissions show that packet switching does not work properly in a
low S/N environment.

Fig. 3.7 BLER versus packet size

.

3.3.1. Discussion of Measurement Results

The result based on the measurements is a recommendation to use FSK-modems in delta
modulated 16 kbps networks for analog data communication. The standard 4-DPSK-modem
is omitted from the recommendation because of its poor BER quality. Standard modems
based on phase differences were not suitable in 16 kbps networks. In fact modems (>1200
bps) standardized by ITU-T using discontinuous phase are problematic in the present delta
modulated voice channels. This result was found after many tests with carefully selected
input levels according to EUROCOM standards and modem specifications.

In most references of delta modulation the slope overload is discussed. The slope overload is
the error of amplitude and phase in a delta modulator output when the output signal does not
follow the input signal. It will be simulated in the next chapter.

In data transmission a PSK modem has rapid phase shifts. A phase shift may have a rapid
increase in amplitude, which a delta modulation system is not able to follow. An 8-bit code
PCM-system has timing pulses and can follow amplitude and phase shifts (adapt to the signal)
better than a one-bit code DM-system. The result is expected because of time varying
slope overload in ADM. Figure 3.6 shows clearly that BER using standard modems in the
analog data transmission over ADM voice channels is in the range 10-3…10-4. BER using
the recommended digital data transmission method with error correction is about 100 times
better according to reference [Eur86].

35


Page 36

Loss of Synchronization

The threshold for loss of synchronization (SL) was 1 in 16, using the data communication
analyzer 2871 of Marconi Instruments Ltd. In the Figure 3.5 made by this analyzer the loss
of synchronization is seen at 14:31:55 and sync gain at 14:31:57. Synchronization was lost in
this case for two seconds. It is a serious problem in data transmission over the ADM-channel
but not a big problem in voice communications. Later the loss of synchronization (SL) was
again monitored now with Navtel Datatest 3. The mean period between the loss of synchronization
was measured with 22 different ten-minute tests. The results were:
- FSK 1200 bps: 1.197 seconds.
- 4-DPSK 2400 bps: 0.985 seconds.
Measurements in another situation with another transmission line gave different results than
in 1993 when the results were about 0.5 seconds for both modem types. Thus these measurements
are unreliable and only indicative. In general to avoid using confusing results a
simulation investigation will be made in the next chapter.

Real Data Rate

Using ADM-channels with analog data modems the loss of synchronism happens very often.
This result was seen as a decrease of the real data transmission speed of FSK to 587 bps and
4-DPSK to 2044 bps. This is an illustrative and suggestive measurement made in 1994. Practical
situation are varying thus to get repeatable, reliable and more general results one will
use simulation methods rather than measurements.

3.3.2 Analysis of Packet Switching

An analysis of packet switching in a network is made using a FSK waveform based on the
recommended measurement results. Figure 3.8 illustrates that packet switching does not
work properly in a bad SNR environment as the probability of retransmission increases with
the block size. The situation found and known in practice and is demonstrated here. In the
example of Figure 3.8 the retransmission phenomena as block error rate (BLER) versus
packet sizes is calculated for the FSK waveform at different low SNR ratios.

36


Page 37

Fig. 3.8 Probability of a FSK retransmission versus packet size [Lal97a]

Formulae (3.1) – (3.3) were used in the evaluation calculations of Figure 3.8. The BER values
correspond with the S/N = 9…12 dB of FSK-waveform. In local area networks (LAN)
the problem is often message delays. A data network modeled with a network level simulator
(Comnet 3) is presented in Figure 3.9.

Fig. 3.9 Example: network model (CACI Comnet 3) [Lal04a]

37


Page 38

Message Delay versus Bit Rate

The simulation results in Figures 3.10- 3.12 show mean and deviation values of message
delays as the effect of different capacity links used in the backbone network. The traffic
bursts are seen in three different cases (600 bps, 9600 bps and 22.5 kbps), in Figure 3.11,
smoothing with the proper channel capacities, in the example case at 22.5 kbps of Figure
3.12. The planning of data channels and network structures are essential in minimizing traffic
delays and bursts. The simulated results in this study motivate to look for better modulation
methods to avoid retransmissions and traffic delays. Comnet 3 is a traffic simulator and
an example of simulator packages, which are discussed in the simulation study in the next
chapter.

Fig. 3.10 Message delay versus channel bit rate [Lal04a]

38


Page 39

Fig. 3.11 Message delay bursts using different backbone networks [Lal04a]

Fig. 3.12 Message delay bursts using a 22.5 kbps backbone network [Lal04a]

39


Page 40

The block error probability may be used in the evaluation of packet transmission. The packet
size especially has to be evaluated. Other parameters used in the evaluation of packet transmission
are:
- Block size.
- Bit error rate.
- S/N.
- Block error probability (probability of retransmission).
In packet transmission some delay problems are eliminated in the proper selection of these
parameters and network topology. Network topology is a large subject of other studies.

In reference [Com92] a formula (3.1) is used for the estimation of the probability P of a correctly
received message.

N

= (1 Pb (3.1)
P )

Where
N = Number of bits in the received message

Pb = Bit error rate. Bit errors are independent variables.

P = Probability of a correctly received message.

The probability of retransmission PB is thus 1-P, formula (3.2).
P

B

= 1 P (3.2)

Where PB = Block error rate or the probability of retransmission.

In reference [Car86] pp. 552-554 a comparison of digital modulation systems give a formula
(3.3), which is used for the estimation of the probability Pe of an OOK or FSK modulation
with envelope detection of a correctly received message.

P
e

γ
2

b

1
= e (3.3)

2

Where γ

b

is the energy-to-noise ratio needed to get a specified error probability per bit Pe .

40

© 2009 OpenThesis.org. All Rights Reserved.