IQdemodulation
IQdemodulation
IQdemodulation
IQ-demodulation
of
RF-data
by
Johan Kirkhorn,
IFBT, NTNU
Table of Contents
1 INTRODUCTION ........................................................................................................................................3
1.1 Abstract...............................................................................................................................................3
1.2 Definitions/Abbreviations/Nomenclature..............................................................................................3
1.3 Referenced Documents ........................................................................................................................3
2 RF SIGNAL ...............................................................................................................................................4
1 Introduction
1.1 Abstract
This document gives an introduction to the IQ-demodulation format of the RF-data stored
from the Vingmed System Five. The document is intended for users of the RF options on the
System Five.
Note that the information given is simplified to present a comprehensive functional overview
of the topic covered, and might not reveal the actual details of the system in full.
1.2 Definitions/Abbreviations/Nomenclature
RF Radio Frequency. The term “RF data” is commonly used to
denote unprocessed data
IQ In-phase Quadrature.
Used to denote the complex format on which the RF data is
stored from the System Five. The IQ demodulation is also
sometimes named Base-band demodulation, Quadrature
demodulation, Complex demodulation etc.
FIR Finite Impulse Response
FFT Fast Fourier Transform
LP Low pass
BP Band pass
A/D Analog to digital
SNR Signal to Noise Ratio
DSP Digital Signal Processor
2 RF signal
Σ
Up to RF-
128 Signal
channels
Modern ultrasound probes (linear- and phased array transducers) consist of several
rectangular elements of piezoelectric material. The piezoelectric elements are capable of
converting varying pressure to electrical signals.
At receive, the analog signal from each individual probe element is first amplified (Analog
Gain) to ensure optimal use of the dynamic range of the Analog to Digital (A/D) converters.
The analog gain factor varies with depth to amplify signals from deep regions most (TGC,
Time Gain Compensation).
The sampled signals are delayed individually to focus the beam to a certain depth and
direction. The delayed signals are weighted to obtain the desired apodization and beam
profile. Finally, the weighted and delayed signals are summed in phase, and this is the RF
signal.
The sampling rate of the RF signal at the output of the beamformer is 20 MHz, and the
resolution is 20 bits. The sampling rate of the A/D converters is 40 MHz with a resolution of
12 bits. 7 bits are added by summing 128 channels (128=27) and the last bit is added when the
sampling rate is reduced from 40 to 20 MHz.
It is also important to note that when acquiring RF-data on the System Five, the Analog Gain
and weights applied to the individual channels are not affected by the Gain knob and TGC
sliders on the front panel. But the overall gain is adjusted if transmit power is changed.
3.1 Introduction
The Nyquist sampling theorem states that to get a unique representation of the frequency
content of a signal, the signal must be sampled at a rate twice the frequency of the highest
frequency component of the signal.
The received RF-signal from an ultrasound transducer is a band-pass signal. The relative
bandwidth of the transducer is usually less than 100%, typically 50-70%. The percentage is
the ratio of the bandwidth to the center frequency of the transducer. The bandwidth is the
frequency range where the sensitivity of the transducer is above a certain level. For one-way
response, this level is usually defined as 3 dB below the level at the most sensitive frequency.
Figure 2 illustrates the frequency spectrum of the received RF signal from a 2.5 MHz probe
with bandwidth B less than 100% of the center frequency. The RF signal is real-valued, which
means that the spectrum for the negative frequencies is a mirrored replica of the spectrum for
the positive frequencies. The sampling frequency (fs) is 20 MHz, meaning that the signal
contain a unique representation of frequencies between 0 and half the sampling rate (10
MHz). The upper limit is usually referred to as the Nyquist limit, or the Nyquist frequency.
Bandwidth
of interest Nyquist
frequency
=
fs /2
B
3.75
f [MHz]
In our example, the transducer is sensitive in a band less than 2.5 MHz wide and centered
around 2.5 MHz. This means that all frequency content of interest lies between 1.25 and 3.75
MHz. Sampling at 20 MHz (as done in the System Five), will therefore be an “overkill” in
terms of amount of data to be transferred and stored. Without loss of information, the
sampling rate could be reduced to about 7.5 MHz. Because the sampling rate in the System
Five is fixed in hardware, this is not easy to do. One could decimate the RF-signal by a factor
2, and achieve a sampling frequency of 10 MHz, which would be an improvement, but not
optimal.
A smarter approach for reducing the amount of data without loosing essential information is
to apply a complex base-band modulation technique with bandwidth reduction known as IQ-
demodulation.
Another issue is, that for suppression of quantization noise during analog to digital
conversion, it is fortunate to keep the sampling rate as high as possible to obtain a better
Signal-to-Noise-Ratio (SNR).
3.2 IQ-demodulation
The IQ-demodulation consists of 3 main steps:
• Down-mixing
• Low-pass filtering
• Decimation
The multiplication with the square root of two is included to preserve the energy in the signal
(explained in section 3.4.)
exp(-i2πfdt) sqrt(2)
RF-signal IQ-signal
LP- Deci-
x RF (t) filter mation x IQ (t)
Down-
mixing
Figure 3. IQ demodulation
3.3 Down-mixing
Down-
mixing
-10 -5 0 10
f [MHz]
Figure 4. Down-mixing
The real valued RF-signal is multiplied (“mixed”) with a complex sinusoid signal:
xIQ(t)=xRF(t) ⋅exp(-i2πfDemod⋅t)
where t is the time along the beam. The relationship between time and distance r is: t=2*r/c.
c is the velocity of sound in human tissue (1540 m/s). The resulting signal xIQ(t) is complex.
Looking at the signal before and after the mixing explains the name “down-mixing”. The
frequency spectrum is actually moved down (to the left) in the frequency plane. After the
down mixing, the resulting signal is complex, and the frequency spectrum is no longer
symmetric about zero.
Because of the relationship between complex exponential functions and sine and cosine
functions,
exp(-iωt)=cos(-ωt)+i⋅sin(-ωt)=cos(ω t)-i⋅sin(ωt)
the down-mixing can be thought of as mixing the RF-signal with two sinusoid signals with
90°phase difference:
cos(2 πfdt)
Re{x(t)}
RF-signal
x(t)
Im{x(t)}
-sin(2 πfdt)
If the demodulation had to be done in hardware, this would be the approach to use. But in the
System Five, the demodulation is done in software using Digital Signal Processors (DSPs),
and the complex exponential is used instead.
The down mixing operation multiplies the RF-signal with a complex vector with unit length,
and the energy content of the signal is not changed.
LP-filter
The filter removes the frequencies stemming from the negative spectrum of the real RF signal,
and the filter removes approximately half of the energy in the signal. In order to preserve the
energy in the signal, the complex signal should be multiplied by the square root of 2.
3.5 Decimation
The Nyquist theorem then states that the sampling frequency can be reduced to twice the cut-
off frequency of the filter without loss of information. Because we have a complex signal, the
bandwidth of the signal equals the complex sampling rate (the complex signal doesn’t have an
ambiguity between positive and negative frequencies).
Figure 7. Decimation
This means that we can reduce the sampling frequency from 20 MHz to 3.33 MHz. 3.33 is the
smallest integer fraction of 20 which is larger than twice the filter cut-off frequency. The
sampling rate is reduced by a factor 6. In practice, the desired decimation is obtained by
keeping every 6th sample and throwing away the rest.
The IQ demodulation preserves the information content in the Band-pass signal, and the
original RF-signal can be reconstructed from the IQ-signal.
The next chapter explains how to reconstruct the RF-signal from the IQ-signal.
The IQ data is written to EchoPAC files with 16 bit signed integer representation of the I and
Q components, giving a total of 32 bits for representation of each sample.
4.1 Interpolation
The first step of the reconstruction, is to increase the sampling rate back to the rate it had prior
to the decimation:
Figure 9. Interpolation
Most signal processing textbooks cover the topic of interpolation/sampling rate conversion, so
the topic will not be covered in deep detail here. The process is divided into two main steps,
zero-padding and low-pass filtering.
4.2 Zero-padding
Zero padding means inserting zeroes in the signal to increase sampling rate. In our case, we
insert 5 zeroes between each signal sample. In the frequency domain, this will be seen as 5
new replicas of the low-pass spectrum, spaced with the original sampling frequency:
LP-filter
NI=round(fS_des/fS_IQ)
The IQ-signal stored from the System Five is decimated from 20MHz by an integer factor
(typically 3 or higher), so the fraction fS_des/fS_IQ should always be an integer. However,
because of finite numerical representation, the fraction should be rounded off to the nearest
integer.
The interpolation filter operates separately on the I and Q components of the IQ-signal. Thus,
the cut-off frequency of the interpolation filter should be set >= 0.5 fS_IQ.
4.4 Up-mixing
To shift the frequency spectrum from the base-band and back to it’s original band, the
interpolated signal is up-mixed.
Up-
m ixing
-10 0 2.5 10
f [MHz]
Where:
t (r)=(StartDepth+(r*DepthIncrement/NI )) / (2*c))
StartDepth The distance from the transducer origin to the start of the IQ
sector measured in meters. See EchoMAT manual for how to extract
parameters from the EchoPAC file.
DepthIncrement Radial sampling interval measured in meters.
StartDepth might be omitted in the expression for t(r) because it only represents a constant
phase shift applied to all samples. The factor 2 in the denominator of the fraction refers to the
fact that the ultrasound pulse travels back and forth from the transducer to the reflecting
target.
RF(r)=sqrt(2)*Re{IQup-mix(r)}
The factor sqrt(2) is included to compensate for the loss of half the energy in the signal when
taking the real value of the complex signal (the energy is assumed to be equally distributed
between the real and complex parts). The resulting real signal has a symmetric frequency
spectrum, as it had in the beginning.
Fundamental Harmonic
frequency frequency
In the figure, the spectrum is shown with two frequency axes. The IQ frequency axis indicates
the frequencies of the spectrum of the IQ signal. The lower axis indicates the true frequencies
from the RF signal. The IQ axis is shifted 2.5 MHz to the left relative to the RF axis. 2.5 MHz
equals the demodulation frequency.
If we want to filter out the second harmonic frequency band, we first down-mix the IQ signal
to get the second harmonic frequency to the zero frequency:
Down-mixing
Observe that the down-mixing will wrap the lower end of the spectrum into the upper end.
Then we apply a low pass filter with cutoff frequency 0.5 MHz, giving a double sided
bandwidth of 1 MHz:
Up-mixing
The routine RECTFREQ (in the READECHO library) performs the filtering described in this
section, using a rectangular frequency response low pass filter. The rectangular filter is
approximated by a FIR filter with Hamming weighted Sinc coefficients. The RECTFREQ
routine requires the SINC and HAMMING functions from the Matlab Signal Processing
Toolbox.
The RECTFREQ function requires specification of filter order. The filter order defines the
steepness of the filter at the cut-off frequencies, and the suppression of the stop-band. The
filter order should be an even number in order to get a symmetric filter.