Sie sind auf Seite 1von 7

A compact Class D RF power amplifier

for mobile nuclear magnetic resonance


systems
Cite as: Rev. Sci. Instrum. 88, 074704 (2017); https://doi.org/10.1063/1.4994734
Submitted: 30 January 2017 . Accepted: 06 July 2017 . Published Online: 25 July 2017

J. Zhen , R. Dykstra , C. Eccles, G. Gouws, and S. Obruchkov

ARTICLES YOU MAY BE INTERESTED IN

A novel approach to estimating the Doppler shift frequency from quadrature mixer output
Review of Scientific Instruments 88, 073503 (2017); https://doi.org/10.1063/1.4991018

Low noise current preamplifier for qPlus sensor deflection signal detection in atomic force
microscopy at room and low temperatures
Review of Scientific Instruments 88, 073702 (2017); https://doi.org/10.1063/1.4993737

Pulsed high magnetic field measurement with a rubidium vapor sensor


Review of Scientific Instruments 88, 073102 (2017); https://doi.org/10.1063/1.4993760

Rev. Sci. Instrum. 88, 074704 (2017); https://doi.org/10.1063/1.4994734 88, 074704

© 2017 Author(s).
REVIEW OF SCIENTIFIC INSTRUMENTS 88, 074704 (2017)

A compact Class D RF power amplifier for mobile nuclear magnetic


resonance systems
J. Zhen,1 R. Dykstra,2 C. Eccles,3 G. Gouws,2 and S. Obruchkov2
1 Centrefor Energy, The University of Western Australia, Perth 6009, Australia
2 Schoolof Engineering and Computer Science, Victoria University of Wellington, Wellington 6012, New Zealand
3 Magritek Ltd., Wellington 6037, New Zealand

(Received 30 January 2017; accepted 6 July 2017; published online 25 July 2017)
A 20 MHz Class D amplifier with an output of 100 W of RF power has been developed. The compact
size printed circuit board area of 50 cm2 and efficiency of 73% make it suitable for mobile nuclear
magnetic resonance (NMR) systems. Test results show that the rise and ring down times of the ampli-
fier are less than 0.2 µs, and it is capable of producing constant amplitude pulses as short as 2 µs.
Experiments using a Carr Purcell Meiboom Gill pulse sequence with a NMR MOUSE sensor confirm
that the Class D amplifier is suitable for mobile NMR applications. Published by AIP Publishing.
[http://dx.doi.org/10.1063/1.4994734]

I. INTRODUCTION filter and converts the rectangular voltage pulses to a smooth


sine wave output.
The physical size of mobile Nuclear Magnetic Resonance Figure 1 shows a voltage switching amplifier where the
(NMR) systems is often dominated by the power amplifier tuned circuit LCR elements are in series. In the case of a cur-
required to excite the NMR probe, usually Class A or AB. Also, rent switching amplifier, the LCR elements are connected in
the efficiency of these linear amplifiers is under 50% in prac- parallel. The operation of both voltage switching and current
tice.1,2 These factors therefore limit the application of mobile switching Class D circuits is the same, the difference being
NMR systems. Class D amplifiers have been used in audio and the current source supply and the type of tuned circuit on the
low frequency NMR applications and easily achieve efficien- output.1,2 In practice, the voltage switching type is preferred
cies as high as 90%.3,4 The challenge is to design such high due to the lower current rating requirements of the MOSFETs,
efficiency amplifiers to operate in the tens of MHz frequency as high voltage MOSFETs are readily available on the market.
range with output powers of 100 W and above. The operating There is also a report 9 that the voltage switching type has a
frequency of Class D amplifiers is limited by the switching slightly higher efficiency at high power when operating at high
capability of MOSFET devices, which require a finite time to frequencies. Thus the voltage switching amplifier has been the
turn on and off, as well the capability of the driver circuit to focus of this research.
rapidly charge and discharge the MOSFET gate capacitance.
With the advances in MOSFET technology, high power and B. Power switching devices and gate driver circuit
high frequency switching MOSFETs are now available on the
market.5,6 Using carefully designed driver methods, it is pos- The main advantage of a Class D amplifier is power effi-
sible to produce Class D amplifiers for NMR applications in ciency. Because the switching transistors (usually MOSFETs)
the tens of MHz frequency range. A Class D amplifier oper- are operated in either conductive or non conductive states
ating with 6 W output power at 4 MHz has previously been only, power dissipation is low as MOSFETs have very low
demonstrated.7 The work here will present an amplifier design on-resistances, RDS(ON) . Class D amplifiers can thus achieve
with an output power of 100 W, operating at 20 MHz which efficiencies of 80% or higher. Recently, power devices have
is suitable to work with a Mobile Universal Surface Explorer become available with fast switching times and low gate capac-
(MOUSE) NMR sensor.8 The performance of the amplifier itance. For example, CSD17313Q2 from Texas Instruments10
is also compared with a commercially available Class A has gate capacitance Cg = 0.45 nF, turn-on rise time tr = 2.8 ns,
amplifier.14 turn-off fall time tf = 1.3 ns, maximum drain/source voltage
rating VDS = 30 V, maximum continuous drain current rating
ID = 19 A, drain/source on resistance RDS(ON) = 24 mΩ, and
II. DESIGN AND PERFORMANCE typical maximum power dissipation of 2.4 W when mounted
on a printed circuit board (PCB).
A. Voltage switching Class D amplifiers
The gate driver is also an important part of the Class D
A simplified Class D circuit with a tuned network out- circuit, as the MOSFETs must be turned on and off quickly
put is shown in Fig. 1. There are two transistors or MOSFET to ensure low switching losses. For that the driver needs to
switches: the one on the top, Q1, is usually referred to as the charge and discharge the gate and parasitic capacitors of the
“high side,” and the other, Q2, is referred to as the “low side.” MOSFET within the switching times of the devices. For a gate
The output from the two switches drives a tank network with drive voltage of 5 V to switch within 1.3 ns, the required current
the load as part of this network. This tank network acts as a is

0034-6748/2017/88(7)/074704/6/$30.00 88, 074704-1 Published by AIP Publishing.


074704-2 Zhen et al. Rev. Sci. Instrum. 88, 074704 (2017)

FIG. 1. Simplified Class D circuit. Note that the wave-


forms are for illustrative purposes only and do not
represent the true shape or timing.

FIG. 2. Gate driver and MOSFET circuit that can pro-


duce a 20 MHz square wave output.

dV   5 see Fig. 3. The dead-time has a direct impact on the efficiency


i=C = 0.45 × 10−9 × = 1.73 A.
dt 1.3 × 10−9 of the amplifier, and it is adjusted to be as short as possible
A gate driver circuit, shown in Fig. 2, uses ISL55110 devices while avoiding shoot-through.12
from Intersil11 to drive a set of MOSFETs in a Class D Figure 4 shows the gate driver inputs and outputs while
arrangement with a target operating frequency of 20 MHz. The driving a complementary pair of MOSFETs with a dead-time
ISL55110 device can provide current pulses of up to 3.5 A and
itself has rise and fall times of 1.5 ns. Note that a floating power
supply is required for the top gate driver (U22) as the common
is connected to the output.
It is very important to ensure that the two MOSFETs are
“on” only one at a time. If both of them are “on” at the same
time, the power supply will be shorted to ground through
the MOSFETs (i.e., a shoot-through), resulting in high cur-
rent flowing through the devices. To avoid this and protect
the devices from damage, a small delay is used between the
switching “off” time of Q1 and the switching “on” time of Q2.
This ensures that Q1 is “off” before Q2 is turned “on” and
vice versa. This delay time is referred to as the “dead-time,”

FIG. 4. Gate driver inputs and outputs at an operating frequency of 20 MHz.


The blue and orange signals are gate input logic signals with 10 ns dead-
time setting, while green and purple are corresponding driver outputs on the
MOSFET gates. The gate driver output signals are not square shaped due to
the effect of the MOSFET gate capacitance. The peak voltage output of the
FIG. 3. Conduction angles of the power devices in the Class D circuit. driver is 10 V, and the full scale time of the plot is 100 ns.
074704-3 Zhen et al. Rev. Sci. Instrum. 88, 074704 (2017)

setting of 10 ns. The driver outputs correspond to the logic


inputs but with typically a 7 ns propagation delay. The output
of the driver, at turn off, returning to ground level indicates the
gate capacitance of the MOSFET being fully discharged and
hence in the off state. The adequate timing margin between
the MOSFETs switching indicates that the circuit is capable
of operating at 20 MHz.

C. Full bridge voltage switching Class D amplifier


The output power requirement of 100 W is considered to
be high for a Class D amplifier working at 20 MHz, therefore
FIG. 6. LT Spice simulation output at 20 MHz. Vp-p = 56 V, power = 125 W
a full bridge (known as an H bridge) configuration was used. RMS into the 3.125 Ω transformer primary.
The full bridge configuration allows for better control of the
amplifier during the end of a transmit stage, as the two bottom
MOSFETs can be used to ground the output nodes to prevent controlled by a KEA 2 spectrometer (Magritek Ltd.,
any oscillations. This is very important for NMR systems as Wellington, New Zealand) which uses a 24 V DC supply.
it affects the rise time and ring down of the RF pulses being The power capability of the Class D circuit and component
generated. A resonant tank was used to convert high power values was found and simulated using LT Spice.13 The induc-
pulses into sinusoidal waveforms, and a transformer (BN-61- tance of the LC tank circuit is the leakage inductance of the
002 material) was used at the output for impedance matching transformer and was measured to be 40 nH. The capacitance
to a 50 Ω load. A schematic diagram of the amplifier out- value was adjusted to be approximately 1.68 nF. The output
put stage is shown in Fig. 5. The circuit is similar to a full waveform from the simulation is shown in Fig. 6. The out-
bridge resonant LC power converter, the difference being that put voltage across the transformer primary has a peak to peak
it does not have the rectifier diodes on the output. Switches S1 value of 56 V, corresponding to 125 W RMS power into the
and S4 are controlled to give the positive output cycles, while 3.125 Ω load. It is the power on the primary side of the trans-
S2 and S3 are used to produce the negative cycle. The reso- former, but there will be power losses in the transformer and
nant tank formed by the LC network was tuned to resonate at the actual power capability would be lower. Even with the
20 MHz. losses, it is expected that at least 100 W will be delivered
To achieve high output power from a Class D configura- to the secondary side of the transformer to drive the 50 Ω
tion with a relatively low supply voltage, the load to be driven load.
has to be of low impedance on the order of few ohms. The A prototype of the design was built on a 4-layer 50 cm2
NMR MOUSE sensor impedance is typically 50 Ω, there- PCB with the finished amplifier weighing just under 120 g.
fore the transformer at the output was chosen to have a 1:4 An existing commercially available 100 W Class A amplifier
turns ratio, thus giving a reflected impedance of 3.125 Ω that is considered somewhat portable and suitable for working
on the primary side. The RF amplifier was designed to be with the NMR MOUSE sensor has a PCB area of 160 cm2 and

FIG. 5. Schematic of a full bridge Class D RF amplifier


output stage.
074704-4 Zhen et al. Rev. Sci. Instrum. 88, 074704 (2017)

The use of such a heat-sink is essential in this case due to the


inefficiency of the Class A amplifier.

D. Performance
The performance of our Class D amplifier was compared
to the existing Class A amplifier. Typical pulse lengths used
with the NMR MOUSE are between 2 µs and 10 µs.8 Figure 8
shows the two amplifiers outputting a short 100 W pulse into
a 50 Ω load. It can be seen that the Class A amplifier takes
about 1 µs to settle, whereas the Class D amplifier reaches the
maximum amplitude after the second cycle and stays constant
for the pulse length. Normally amplifiers with a resonant tank
output have the problem of long rise and ring-down times; this
problem was overcome by using a lower Q tank circuit and the
intelligent switching of the MOSFETs where at the end of the
pulse, the low side MOSFETs are turned on to reduce the ring
down time.
One of the NMR pulse sequences used with the NMR
MOUSE is the spin echo sequence15 and consists of two RF
pulses with the first pulse, a 90° pulse, being half the energy
FIG. 7. Existing Class A amplifier (a) with large heat-sink and the new Class
D amplifier (b). The Class A amplifier has a PCB area of 120 cm2 and weighs of the later second pulse known as the 180° pulse. The 180°
480 g; the Class D amplifier has a PCB area of 50 cm2 and weighs 120 g. pulse is either double the amplitude or double the length of
the 90° pulse. With classic (Class A) amplifier systems, the
weighs 480 g.14 A photo comparison of the two amplifiers is
shown in Fig. 7. One big difference between the amplifiers is
the heat-sink (bottom side of PCB) on the Class A amplifier.

FIG. 9. Spin echo sequence RF pulses from Class A amplifier (a) and Class
D amplifier (b) into a 50 Ω load. Note in this case, the Class D amplifier output
power level is fixed so pulse modulation is performed by altering the pulse
FIG. 8. 100 W pulses from Class A amplifier (a) and Class D amplifier (b) length. The Class D trace also shows the amplifier gating control signal that
into a 50 Ω load. The same input waveform was used in both cases. is typically applied to NMR pulse amplifiers.
074704-5 Zhen et al. Rev. Sci. Instrum. 88, 074704 (2017)

double amplitude method is used as it provides constant excita- TABLE I. Class A and D amplifier parameters including power and efficiency
tion bandwidth; however, for the Class D amplifier, the double measurements.
pulse length method is easier and is therefore used. Amplitude Parameter Magritek Class A Home built Class D
modulation using a Class D amplifier could be provided by
either modulating the conduction time of the transistors or by Rated DC supply voltage (V) 24 24
changing the bridge voltage and this will be the focus of some Maximum output power (W) 100 100
future work. Input impedance (Ω) 50 50
A spin echo pulse output from the two amplifiers is shown Output impedance (Ω) 50 50
in Fig. 9. It is interesting to note that the Class A amplifier is Input drive level (mW) <=1 1
still settling before the end of the pulse, whereas the Class D Bandwidth (MHz) 1-50 16-18
amplifier pulses show rapid rise time and very constant ampli- Module size (PCB area) (cm2 ) 160 50
tude even when doubling the pulse length. The Transistor Module weight (g) 480 120
Transistor Logic (TTL) pulse on the same figure shows the Vin (V)a 23.3 23.3
turn on and off times of the RF pulse. Iin (A)b 2.08 0.6
Another useful NMR pulse sequence for the MOUSE is Vout,p-p (V)c 190 202
the Carr Purcell Meiboom Gill (CPMG) pulse sequence16,17 η (%)d 19 73
which generates multiple echoes by using more than one 180° a V is the actual DC supply voltage to the amplifier.
in
pulse. This experiment is typically used to measure the sample b I is the difference in input current between the amplifier in the idle state and running
in
relaxation time T 2 . The efficiency of the two amplifiers was state.
cV
compared using their average input and output power over a out,pk-pk is the peak to peak output voltage measured on the 50 Ω load.
d η is the power efficiency, calculated using peak output RMS power at the duty cycle of
long CPMG pulse sequence consisting of 128°, 180° pulses at 10% verses DC input power.
10% duty cycle. A current probe was used to measure the cur-
rent drawn from the power supply of the entire spectrometer
system while the amplifier was running. This was then com-
pared to the steady state current demand when the amplifier
was not operational, the difference being the power required
for the RF generation (Fig. 10). It should be noted that both
amplifiers were expected to have similar and small standby cur-
rents, therefore justifying a relative efficiency measurement as
a good indicator of amplifier absolute efficiency. The DC input
voltage was measured to be 23.3 V, and the same 50 Ω load
was used to test both amplifiers. The corresponding RF gen-
erating current component for the Class A amplifier is 2.08 A
and 0.6 A for Class D. The RF output voltage for the Class A is
190 Vp-p and 202 Vp-p for the Class D. Based on those values,
the efficiency of the Class A amplifier was calculated to be
19% and 73% for Class D. A summary of the measurement
parameters as well as other amplifier parameters is shown in
Table I.

III. NMR MEASUREMENTS


The ultimate test of the amplifiers is the NMR experiments
themselves. The CPMG experiment is sensitive to phase insta-
bilities during the RF pulses as well as the pulse ring down, so
it is a good test to verify the suitability of the Class D ampli-
fier. Both amplifiers were used with the MOUSE sensor to
perform the same relaxation experiments on a small sample
of rubber. Figure 11 shows Class A and Class D spin echo
and CPMG experiment results. The spin echo results were
obtained from a phase-cycled data-averaged sequence of 16
scans. The two results being similar is an indication of the two
amplifiers behaving in a similar fashion over the duration of
FIG. 10. Power and efficiency measurements for Class A amplifier (a) and the 16 scans. The CPMG results were also obtained from the
Class D amplifier (b) driving into a 50 Ω load and performing a CPMG pulse average of 16 phase cycled scans. This time the pulse sequence
sequence. The top traces in the figures (blue) are from a current probe set for an
is much longer, with 128 echoes forming in each sequence;
output of 100 mV/A. Note that the step response shape of the current demand
is largely due to the properties of the power supplies and filter capacitors used this is shown as the raw CPMG echo data in the figures. The
and is not an important characteristic of the amplifier. similar T 2 decay constant of approximately 5 ms and similar
074704-6 Zhen et al. Rev. Sci. Instrum. 88, 074704 (2017)

FIG. 11. Spin echo results from the Class A amplifier (a)
and Class D amplifier (c). CPMG results from the Class
A amplifier (b) with T2 determined to be 5.5 ms and the
Class D amplifier (d) with T2 = 4.7 ms.

amplitudes suggest that the behavior of the Class D amplifier multiple primaries in a single transformer. This would allow
is close to that of the Class A amplifier. the doubling of the output power with no expected increase in
the circuit board area.
1 M. Albulet, RF Power Amplifiers (Noble Publishing Corporation, Atlanta,
IV. CONCLUSION
GA, 2001).
2 M. K. Kazimierczuk, RF Power Amplifiers (John Wiley & Sons, West
The similarity of the NMR test results between the Class
Sussex, 2008).
A and Class D amplifiers suggests that the latter is capable 3 H. Koizumi, T. Suetsugu et al., “Class DE high-efficiency tuned power
of producing the output power and pulse sequences necessary amplifier,” IEEE Trans. Circuits Syst. I: Fundam. Theory Appl. 43(1), 51
for NMR experiments. The efficiency of 73% at the operating (1996).
4 T. Hopper, Y. Q. Song et al., “Low-frequency NMR with a non-resonant
frequency of 20 MHz with 100 W output power, combined
circuit,” J. Magn. Reson. 210, 69–74 (2011).
with the fast rise time and short turn off time, is a significant 5 N. Dye, H. Granberg et al., Radio Frequency Transistors—Principles and
improvement when compared with the existing Class A ampli- Practical Applications, 2nd ed. (Newnes, USA, 2001).
fier used with the NMR MOUSE sensor. However, it should 6 F. Schwierz and J. J. Liou, “Development of RF transistors: A historical

be noted that the Class D amplifier described here is a tuned prospect,” in Proceedings of International Conference on Solid-State and
Integrated-Circuit Technology (IEEE, 2001), Vol. 2, pp. 1314–1319.
amplifier and therefore does not have the benefit of the Class 7 X. Zhang, N. Schemm et al., “A novel power amplification scheme for
A amplifier of wide bandwidth and the ability to operate at nuclear magnetic resonance/nuclear quadrupole resonance systems,” Rev.
multiple frequencies. Nevertheless, the compact size and light Sci. Instrum. 82, 034707 (2011).
8 G. Eidmann, R. Savelsberg, P. Blümler, and B. Blümich, “The NMR
weight make the Class D amplifier a key enabler for portable
MOUSE, a mobile universal surface explorer,” J. Magn. Reson., Ser. A
NMR systems. 122, 104–109 (1996).
Some applications will require more than 100 W of 9 W. J. Chudobiak and D. F. Page, “Frequency and power limitations of Class-

RF power and or greater than 20 MHz operating frequency. D transistor amplifiers,” IEEE J. Solid-State Circuits 4(1), 25–37 (1969).
10 See http://www.ti.com/product/CSD17313Q2 for Texas Instruments.
There are several possible solutions to this need. First, new 11 See http://www.intersil.com/en/products/amplifiers-and-buffers/all-amplifi
power transistors with improved voltage and current handling ers/powerfet-ccd-drivers/ISL55110.html for Intersil.
capability as well as reduced gate capacitance are continu- 12 A. Grebennikov and N. O. Sokal, Switchmode RF Power Amplifiers

ally entering the market so it is expected that higher output (Elsevier, Inc., MA, 2007).
13 See http://www.linear.com/designtools/software/ for LTSpice.
powers and operating frequencies will be obtained in the 14 See http://www.magritek.com for Class A RF amplifier produced by
future by simply replacing the power devices and increas- Magritek Ltd.
ing the supply voltage. Second, several of the described 15 E. L. Hahn, “Spin echoes,” Phys. Rev. 80(4), 580–594 (1950).
16 S. Meiboom and D. Gill, “Modified spin-echo method for measuring nuclear
100 W modules could have their output powers combined
relaxation times,” Rev. Sci. Instrum. 29, 688–691 (1958).
through a power combiner such as a lumped element quar- 17 H. Y. Carr and E. M. Purcell, “Effects of diffusion on free precession
ter wave (Wilkinson) power combiner. Third, a new design in nuclear magnetic resonance experiments,” Phys. Rev. 94(3), 630–638
could be implemented where multiple driver stages drive (1954).