RF TRANSMITTER
This invention relates to RF transmitters.
The invention particularly relates to RF transmitters for generating suppressed carrier
RF signals from in-phase and quadrature baseband input signals.
To prevent energy being carried over into sidebands which would contravene
broadcasting regulations, one technique used is for the transmitter to employ linear
power amplifiers.
Accordingly, a feedback loop means is used so that the in-phase and quadrature input
signals are compared with signals derived from the output of the power amplifier
means, de-modulated using quadrature RF reference signals, to produce respective in-
phase and quadrature feedback signals, which are used to produce error signals, and
the error signals are modulated onto respective RF quadrature reference signals and
combined to produce the suppressed carrier RF signal for input to the power amplifier
means such that in the process the feedback is responsible for pre-distorting the input
to reduce non-linearities in the power amplifier means.
Unfortunately this can only correct for non-linearities in the power amplifier and
modulator means and not for errors in the feedback loop itself. It is found that such
errors are actually responsible for introducing DC offsets into the feedback loop which
are then responsible for re-introducing a component of the carrier into the output.
It has been proposed to reduce this DC offset error in an iterative manner by making
a series of adjustments to the DC input level of the in-phase and quadrature input
signals and maintaining those which were responsible for reducing the amount of re-
introduced carrier. However, it was found that this procedure took a relatively long
time in order to be effective.
The invention provides an RF transmitter for generating a suppressed carrier RF signal
from in-phase and quadrature baseband input signals, comprising power amplifier
means, feedback loop means arranged so that the in-phase and quadrature input signals
are compared with respective signals demodulated using quadrature RF reference
signals from signals derived from the output of the power amplifier means, to produce
respective in-phase and quadrature error signals, means for modulating the error signals
onto respective quadrature RF reference signals and for combining them to produce a
suppressed carrier RF signal for input to the power amplifier means, the input thus
being pre-distorted to reduce non-linearities in the power amplifier means, and means
for applying a calibration DC component to the in-phase and quadrature input signals
based on measurements of signals derived from the output of the power amplifier
means when a known DC offset is injected into the in-phase and quadrature input
signals, to reduce carrier breakthrough.
The calibration can be performed relatively rapidly, because the pre-distorting DC
component is calculated from the measurements made during a defined period.
The values can be measured by using a detector such as a diode detector to sense a
signal derived from the output of the power amplifier means in order to detect the
envelope amplitude variation corresponding to the transmitted output, and correlation
means or filtering means can be employed to quantify the amount of the amplitude
variation which relates to the carrier being re- introduced.
An RF transmitter constructed in accordance with the invention will now be described,
by way of example, with reference to the accompanying drawings, in which:
Figure 1 is a block diagram of a base station for private mobile radio system, in which
base station the transmitter is incorporated;
Figure 2 is a block diagram of the RF transmitter;
Figure 3 is a block diagram of a Cartesian Loop RF Stage of the transmitter;
Figure 4 is a phase diagram showing the possible phase changes of the transmitted
carrier;
Figure 5 is a spectral diagram showing the generation of a tone offset with respect to
the carrier frequency;
Figure 6 is a vector diagram showing the vectors of the tone illustrated in Figure 5 and
a small component at carrier frequency;
Figure 7 is a spectral diagram showing the two vectors shown in Figure 6;
Figure 8 shows the construction of the detector 31 of Figure 3;
Figure 9 illustrates signals M0. Mj. M2, M3, M4 used to reduce the breakthrough of a
component at the carrier frequency;
Figure 10 is a phase diagram illustrating DC offsets applied to the I input of the
Cartesian Loop RF Stage;
Figure 11 is a phase diagram illustrating the addition of DC offsets to the Q input of
the Cartesian Loop RF Stage; and
Figure 12 illustrates the calibration process.
Referring to Figure 1, the RF transmitter forms part of a base station of a digital
private mobile radio network. The network typically consists of base stations, each
connected by land lines to switching centres, which interface with the national
telephone system. The users of the system employ mobile stations which communicate
with their nearest base station. Two mobile stations which are in the range of the same
base station may communicate via that base station. Mobile stations in the range of
different base stations may communicate via those base stations and an intermediate
switching centre. In addition, mobile stations can, in certain circumstances,
communicate with other mobile stations. The system is suited to users such as the
emergency services and public transport. The messages are communicated in digital
form so that a range of facilities including voice, circuit mode data, short data
messages and packet mode services are possible.
Referring to Figure 1 , the base station consists of a site controller 1 , a digital signal
processor (DSP) 2, a transmitter output stage 3 and a receiver 4, together with
respective synthesisers 5 and 6. The site controller is the highest level of control
within the base station. Its function is to control the supervisory functions of the base station, including establishing, maintaining and terminating connections to the DSP 2,
which includes typically carrier processor cards. Each carrier processor card processes
signals for transmission and reception at one carrier frequency. A signal received from
a mobile station is demodulated in a carrier processor card and decoded at the receiver
4, whereupon it is retransmitted via a carrier processor card and I (in-phase) and Q
(quadrature) channels are output from the carrier processor card to the transmitter
output stage 3. A separate transmitter is associated with each carrier processor card
(only one transmitter is shown), and the respective synthesiser 5 generates the local
oscillator. Similarly there will be a separate receiver 4 for each carrier processor card,
and a respective synthesiser unit 6.
The transmission is time division multiple access (TDM A), and there are four time
slots per carrier so that, for each carrier processor card, four communication channels
can be maintained. It is necessary, however, to reserve one communication channel
for control purposes.
The modulation performed in the DSP 2 of Figure 1 is differential quadrature phase
shift keying (DQPSK). What this means is that, for each consecutive pair of input data
bits (each dibit) to the modulator 7 of Figure 2. a change is made in the phase of the
transmitted carrier. Referring to Figure 4. which represents in simplified form the
possible phases of the transmitted carrier, the phase can be only one of the eight phase
angles shown ie. 0, ± π/4, + π/2, ± 3π/4, π . Further, each dibit can change the
transmitted carrier only by ± π/4, or ± 3π/4. The actual modulation of the I and Q
channels onto the carrier takes place in the Cartesian Loop RF Stage (the transmitter
output stage). Each dibit input into the modulator 7 produces an output on each of the
I and Q channels. If the position was (which it is not) that the I and Q channels
maintained a constant phase, it can be visualised how successive dibits could produce,
say, on the I channels, successive digital values as to represent a sinusoid and, on the
Q channel similar values but 90° out of phase. In fact, the modulation is differential
phase shift keying, which means that changes in phase of the transmitted carrier
represent the information bits. For example, if the input to the modulator consisted of
successive pairs of dibits each representing 1, 1, the values output by the modulator 7
on each channel corresponding to each input digit could be responsible for advancing
the phase of the I and the Q channel each by 45°. This would be equivalent to
advancing by 45° along the sequence of digital values representing a sinusoid referred
to above. Because the filtering which takes place in the I and Q channels, the phase
changes linearly with time rather than moving in steps. Thus in reality each straight
line in Figure 4 would actually be curved: one curved line is shown as an example.
This linear phase change is manifested in a frequency shift from the original
unmodulated carrier frequency fc. The frequency offset is proportional to the bit rate
of the incoming data string. The gross data rate in the DSP 2 is 36k bits per second,
which gives a rate for the dibits of 18k symbols per second. Since eight π/4 transitions
represents one cycle, the symbol rate results in a shift of the carrier frequency of
2.25kHz.
Referring to Figure 2, the samples being output from the modulator 7 are I and Q
baseband samples and these are fed into channel filters 8, 9 which are root raised
cosine filters implemented using a linear phase digital Finite Impulse Response (FIR)
filter. The digital samples are then converted into analogue signals by means of two
12 bit DACs 10, 11 for the I and Q channels. The analogue signals are then fed into
anti-aliasing filters 12, 13 to remove the sampling frequency component of 144kHz and
fed into the Cartesian Loop RF Stage 3.
The circuit elements 7-13 may be provided by dedicated hardware, but are more
conveniently carried out by the DSP 2.
The analogue signals entering the Cartesian Loop RF Stage on the I and Q channels are
baseband signals, and are modulated on to a carrier in the Cartesian Loop RF Stage,
described in more detail now with reference to Figure 3. The modulation is double
sideband, suppressed carrier (DSB-SC). The I channel is modulated by an in-phase
local oscillator via mixer 14, and the Q channel is modulated by a quadrature local
oscillator via mixer 15. Quadrature phase shift device 16 produces the signals from an
input from the local oscillator from synthesiser 5 (Figure 1) operating at carrier
frequency (in this case 396.5MHz). The I and Q signals thus modulated onto
quadrature carriers are then combined at adder 17 and a DSB-SC waveform results.
Figure 5 shows the spectrum of the transmitted signal, and it will be noted that no
component is shown at the carrier frequency fc.
The Cartesian Loop RF Stage includes a 25W class AB power amplifier illustrated by linear amplifiers 18 to 21 in cascade.
It is necessary to correct for non-linearities in the amplifier chain 18-21, and this is
done by means of the (negative feedback) Cartesian Loop.
A probe 22 picks off a part of the output signal and feeds it back to the input I and Q
channels. Of course, the signal picked off is modulated and so must be demodulated,
again using quadrature signals derived from the same local oscillator by means of
quadrature phase shift device 23 and mixers 24 and 25 which act on part of the
feedback signal which is split at splitter 26. The demodulated I and Q channels are fed back to nodes 27 and 28, where the feedback signals are subtracted from the input
signals. If there is sufficient gain around the loop, the feedback signal will strive to
track closely the input signal, so that the signal leaving the nodes 27 and 28 and being
fed to the frequency changes 14 to 17 are error signals and the input to the amplifiers
29 and 30 behave as virtual earths.
It will thus be seen that the Cartesian Loop is a closed loop system where a part of the
transmitted signal is fed back and compared with the input to produce an error signal.
The error signal is then used to pre-distort the input to the power amplifier so that a
linear response is achieved.
The Cartesian Loop only cancels out distortions if the feedback loop is perfect.
Unfortunately, this is not the case, and in fact both the splitter 26 and the mixers 24,
25 introduce a small DC offset into the signal picked off from the RF output 22. The
DC offset is responsible for introducing into the transmitted output a small component
at the carrier frequency fc (Figure 7), since the local oscillator mixed with DC produces
the carrier.
The result of this so-called carrier breakthrough can be considered by reference to
Figure 5, 6 and 7. It will be remembered that the large vector in Figure 7 is the tone
generated at the frequency offset by 2.25kHz from the carrier. The small vector is the
re-introduced carrier component. The carrier component is therefore modulated by a
much larger component at 2.25kHz. It is easier to visualise this by referring the
modulation to the tone, in which case the carrier modulates this tone at 2.25kHz. The
tone is thus amplitude modulated at 2.25kHz.
Detector 31 picks off the feedback signal using probe 32 and detects the envelope
amplitude of the signal. The detector 31 may be configured like the diode detector
shown in Figure 8. The resistor R2 and capacitor Cl provide the necessary time
constant to detect amplitude variations in the envelope of the amplitude modulated
(AM) wave.
The carrier breakthrough causes a ripple at 2.25kHz, and a correlation is performed by
means not shown to detect the amplitude of that ripple.
The ripple detected is a combination of the I and Q channels, and may be represented
by a vector M0 at 2.25kHz. In order to achieve cancellation of the DC offset
introduced in the negative feedback path, it is necessary to calculate this vector M0.
The vector M0 has been illustrated in Figures 10 and 11 and is a complex quantity.
The correlations at 2.25kHz are performed using quadrature waveforms to detect the
I and Q components of vector M0. The vectors Idc and Q^ must be calculated in order
to cancel the DC offset which brings vector M0 into existence. They cannot be worked
out from a knowledge of M0 alone, since they appear at different parts of the circuit
and the relation between them is not known.
The Applicants have previously attempted to reduce the vector M0 in an iterative
manner, by making changes of the DC level in the feedback path and then noting
whether M0 was increased or reduced. Nevertheless, this is a very time-consuming
process.
In accordance with the invention, a calculation is made after five measurements in
order to obtain a measure of the ripple vector M0. In each of five time slots, an initial
period is allowed to elapse before measurements are made to allow transients to decay.
Then the correlation is performed with quadrature 2.25kHz waveforms. In the first
measurement period the I and Q components of vector M0 are calculated. In the second
measurement period ("Measure 1") a positive offset current is generated by DAC 33
and injected into the I input at node 27, and the ripple is measured to calculate vector
Mx. In the next measurement period ("Measure 2") a negative offset of the same value
is generated by DAC 33 and is subtracted from the I channel, and the corresponding
vector M2 is measured. Since the positive DC current offset (+ Δdc) and the negative
DC current offset (- Δdc) can be plotted on the vector diagram 10 and are applied at
the same point in the circuit at which the error Idc is to be cancelled, the unknown
quantity Idc can now be calculated.
In the next measurement slot ("Measure 3"), a positive offset current is added to the Q
channel at node 28 by means of DAC 33 and vector M3 representing the 2.25kHz
component is measured, and in the final measuring period ("Measure 4") a negative DC
offset current is added to node 28 by means of DAC 33, and the ripple M4 is measured.
In the same way, it is then possible to calculate Qdc.
The calculation of Idc and Qdc is performed after "Measure 4" in the DSP 2.
DSP can now, via DAC 33 inject Idc and Qdc current offsets into respective nodes 27,
28, and it is found that this will reduce the 2.25kHz carrier breakthrough component
significantiy. It does not reduce it all in one go, and, but this can be reduced to a low
level in a matter of three of four such adjustments.
Of course the correction is advantageously performed on a continuous basis. The
telecommunication time slots last 14.2 μs, and the whole measurement and calculation
process illustrated in Figure 12 can take place in one time slot.
It will be appreciated that the injected currents ± Δdc are such as to increase and
decrease the DC level at the corresponding node ie. it could be that after injection of
+ Δdc, the DC level moves from one negative level to another, less negative DC level.
The correlation process is performed in DSP 2 and, to this end, an ADC 34 is
provided.
Of course variations may be made without departing from the scope of the invention.
Thus, the invention is not restricted to base stations of private mobile radio networks,
and is also applicable to transmitters with Cartesian Loops in switching centres, or to
transmitters with Cartesian Loops used in base stations or switching centres of other
mobile radio systems or, indeed, more generally to any RF transmitter employing a
Cartesian Loop.