ACE9030
filtered to nearer telephone bandwidth by an on-chip amplifier
with off-chip feedback components.
rapidly increase relative to the I.F. signal and so avoid low
frequency beats, when the system could sit in the state where
a steady part of one cycle is compared with a steady part of
another for a long period of time and so give no output.
The accuracy of the delay is not important as a small error
will only give a D.C. offset in the output but the delay must be
consistant to avoid adding modulation to the output so in the
ACE9030 it is derived from the crystal frequency.
The I.F. signal is digitised at a rate set by the crystal in use
and by the divider D in figure 18 and so will be in the range
4·267 to 7·680 MHz. These rates are all greater than the
maximum audio frequency of 3·4 kHz by a factor of at least
1254 ( which is over 210 ) and so the quantisation allows better
than 63 dB signal to noise ratio in the final audio, even though
only single bit quantising is used. The I.F. is oversampled by
a much smaller ratio and so will have a smaller signal to noise
ratio if measured in its total bandwidth, but this bandwidth is
reduced in the demodulation process to give a good audio
signal to noise ratio in the system.
The sampling rate must not be a harmonic of the I.F., or
very close to one, to prevent the sampling phase becoming
synchronised to the signal and so missing all edges, leading
to the modulation being lost for long periods of time at the beat
frequency (a 14·85 MHz crystal cannot be used in D = 3 mode
with an I.F. at 450 kHz as 14·85 MHz ÷ 3 is 4·95 MHz which is
11 x 450 kHz). It cannot be assumed that a sampling rate
greater than 4 MHz always meets the Nyquist criterion for the
I.F. signal at nominally 450 or 455 kHz because the input
signalisoftenasquarewavefromalimitingamplifierandifnot
is converted to a switching logic signal in the Schmitt trigger
input buffer giving many significant harmonics. The modula-
tion deviation is up to 14·5 kHz and is multiplied by the
harmonicnumbertogiveincreasinglywidedeviationsuchthat
the spectrum eventually becomes continuous, but at a low
level, for the very high (e.g.17th or above) harmonics. A
sampling rate of a few MHz will then retain all required
information and allow distortion free demodulation but is
undersampling in Nyquist terms so aliasing effects must be
avoided by choosing a frequency separated from the nearest
harmonicoftheI.F.byatleasttwicethemodulationfrequency.
Allcombinationsgivenintables2and3cansafelybeusedbut
care is needed if a different crystal or I.F. is required. For
example, a 14·4 MHz crystal cannot be used in ÷ 2 mode with
a 450 kHz I.F. but÷ 3 can be used and with an M of 40 will give
a delay of 3·75 I.F. cycles and alias-free demodulation.
Sampling the I.F. signal at a rate of only 9·3 to 16·9 times
the I.F. will remove the fine detail of the modulation from each
individualcycleoftheI.F.butthemodulationbandwidthisvery
low (both speech and tones) compared to this sampling rate
so the information will be preserved as infrequent whole
sample steps, which when averaged over many samples will
show the correct modulation.
To power down the discriminator a Normal command can
be used:
DATA1
DATA2
DATA3
xxxxxxxx
01 D5 xxxxx
xxxxxxxx
where the discriminator is powered down if DATA2:D5 is LOW
or is set active if DATA2:D5 is HIGH.
The values of the programmable constants D and M are
set by a Set-up command, which can also use DATA1 bit D5
for the lock logic filter period and DATA1 bits D2, D1, D0 for the
OSC8 mode programming:
DATA1
DATA2
DATA3
D7 D6 xxxxxx
10xxxxxx
xx1xxx00
The two control bits D7, D6 set the values for D and M as
in table 2. From this table of frequencies and division ratios it
is possible to calculate the length of the delay M in terms of
cycles of the input I.F. to understand the discrimination proc-
ess shown in table 3.
It can be seen that a 12·8 or 15·36 MHz crystal will give a
delay of a few whole cycles plus or minus one quarter cycle to
a very good accuracy and that a 14·85 MHz crystal similarly
gives some whole cycles plus or minus an odd third of a cycle.
Thesenon-integerdelaysareneededbecausethedelays
are not locked to the I.F. input on AFCIN, and to get a
demodulated output the comparisons must include an edge
time, at least for some samples. The odd quarter or third of a
cycle ensures that the phase of the start of the delay time will
To explain the operation of the discriminator an example
diagram of the sampling points and the comparison delay is
given in figure 19, with the effect of modulation on the input
shown by fine lines. The increasing separation of these dotted
DATA1 bit D7
DATA1 bit D6
Set D
Set M
39
40
37
38
Intended I.F.
450 kHz
Intended Crystal
12·8 or 14·85 MHz
12·8 or 14·85 MHz
15·36 MHz
0
1
0
1
0
0
1
1
2
3
3
2
455 kHz
450 kHz
455 kHz
15·36 MHz
Table 2
D7, D6
0, 0
0, 0
1, 0
1, 0
0, 1
1, 1
D
2
2
3
3
3
2
M
Crystal Freq. Sampling Rate
I.F.
Delay as I.F. cycles
39
39
40
40
37
38
12·80 MHz
14·85 MHz
12·80 MHz
14·85 MHz
15·36 MHz
15·36 MHz
6·400 MHz
7·425 MHz
4·267 MHz
4·950 MHz
5·120 MHz
7·680 MHz
450 kHz
450 kHz
455 kHz
455 kHz
450 kHz
455 kHz
2·742
2·363
4·266
3·677
3·252
2·251
Table 3
21
MITEL [ MITEL NETWORKS CORPORATION ]
