Design and test of a 434 MHz multi-channel amplifier system for targeted hyperthermia applicators

Abstract

Purpose: For our head-and-neck hyperthermia (HT) applicator, an amplifier system with full amplitude and phase-control to deliver the radio-frequency signals, was not available. We therefore designed and tested a 433.92 MHz multi-channel amplifier system.

System description: The design consists of a direct digital synthesizer (DDS) system that generates 12 phase-controlled coherent 433.92 MHz signals, which are amplified to maximum 200 W output per channel. Directional couplers are placed at the amplifiers to couple a small portion of both forward and reflected signals to gain-and-phase detectors. The power setting is applied with a resolution of 2 W and for the phase it is 0.1°. The channels are sequentially sampled at 100 Hz per channel.

Methods: We tested the performance of the designed amplifier system by measuring the RF spectrum, power and phase accuracy, and by characterising the feedback control by using highly accurate power and phase meters.

Results: The spurious emission is less than 60 dBc and the first two harmonic frequencies are suppressed more than 45 dB. The measurement accuracy for the power (±5%) is valid for at least 20 days after calibration and for the phase (±5°) it is valid for at least 2 months.

Conclusions: The amplifier system operates according to our design criteria to support targeted HT. It can be used for both our in-house developed superficial and head-and-neck HT applicators or any other HT applicator that works on the same frequency of 433.92 MHz.

• phased array
• amplifier system
• hyperthermia applicator
• feedback control
• power accuracy
• phase accuracy

Introduction

Several clinical Phase-III trials have demonstrated that addition of Hyperthermia (HT) to radiotherapy results in improved clinical outcome. Good examples are: the long-term survival for advanced cervical cancer is doubled [1], [2] and also the local control rate is doubled for recurrent breast cancer in areas irradiated earlier [3]. Because the clinical outcome is related to the thermal dose [4], [5], a controlled delivery of heat in the tumour is required.

Heat is often applied using electromagnetic (EM) radiation generated by either a single antenna or antenna arrays. To ensure the treatment quality, the thermal dose delivery should preferably be verified by accurately measuring temperatures at a high-spatial resolution [6]. In the clinic, however, interstitial thermometry provides only sparse data. Non-invasive thermometry (NIT) by MRI provides a high resolution, but is not very accurate (±0.5°C) [7], [8], nor always economically or technically feasible. We therefore focus our efforts on enhancing treatment quality by improving the accuracy of the specific absorption rate (SAR) deposition.

We control the SAR by characterising the radiation patterns of our applicators [9–11] and by using state-of-the-art HT planning tools [12]. We use a segmentation tool iSeg (SPEAG, Switzerland) to create tissue-specific 3D patient models from either CT or MRI data. Next, we use an EM simulation tool SEMCAD X (Speag, Switzerland) to calculate the EM fields of our applicators and to optimise the SAR distribution. As a result, a set of antenna phases and amplitudes are prescribed, which have to be transferred to the clinical applicator. Therefore, amplifier systems are needed that accurately deliver power and phase-controlled radio-frequency (RF) signals to the antennas. In this article, we address the amplifier system that was designed for our recently developed head-and-neck applicator.

The accuracy of RF signals is generally influenced by external factors such as (1) EM interference, (2) EM coupling between antennas and (3) impedance mis-matching causing high reflected powers. These factors can cause instabilities and inaccuracies in the RF signals delivered, often depending on the frequency, amplifier gain and temperature of operation. Hence, accurate control of the SAR delivery to the patient requires not only well-designed antennas but also amplifier systems that accurately provide the RF signals.

At the Daniel den Hoed Cancer Center, we use Lucite Cone Applicators (LCA) for superficial HT [13], [14], and just recently we introduced the HYPERcollar [15], [16] for head-and-neck HT. Both of these in-house developed applicators operate at 433.92 MHz and an off-the-shelf commercial phase-controlled amplifier system was not available. For HT applications that work on other frequencies, such amplifier systems have been developed in-house as well [17–20]. An exception is the BSD2000 system (BSD Medical, USA), which is widely used to apply HT in the pelvic area. The BSD2000 operates at frequencies in the range of 60–120 MHz. Several groups have improved the quality of this system, introducing measurement and feedback control loops to verify the powers and phases of the RF signals delivered [21–24].

Since this commercial system does not operate at 433.92 MHz, we needed a new multi-channel amplifier system for the LCA and HYPERcollar applicators. We therefore designed a multi-channel 433.92 MHz amplifier system to control the RF signals for both the LCA system and the HYPERcollar. In this work, we describe the design and establish the accuracy of the power and phase control by using an accurate reference-measurement system.

Design criteria

The ultimate goal of the amplifier system is to apply the RF signals to the LCAs and HYPERcollar, as prescribed by the HT planning tools. In reality, however, power and phase errors are made in the signals delivered which cause suboptimal SAR distributions within the patient. The task addressed in this article is to minimise these errors by designing an highly accurate amplifier system.

The amplifier system will be used for both the LCA in superficial HT and the HYPERcollar in head-and-neck HT, which implies specific requirements. In superficial HT treatments, the power per LCA is increased in steps of 5 W to a maximum of 150 W. Each LCA has a mechanical ‘tuning’ pin to match the antenna impedance to the patient. This pre-treatment tuning procedure requires fast measurements of the reflection coefficient, preferably five samples per second.

For the recently developed HYPERcollar we have more specific requirements, since we use a phased-array which is used to create interference patterns within the patient. The ‘total’ accuracy of the focus-position in the clinical configuration should be ±5 mm. This accuracy can be influenced by errors in, e.g. the segmentation, EM simulation, positioning and finally in the transfer from planned RF signals to the actually applied signals. Therefore, we allow for maximum ±1 mm focus-position error due to inaccurate RF-signals. To estimate the phase accuracy required, we considered an analytical model to predict the influence of phase errors on the SAR focus position. This analytical model consists of a cylindrical configuration of monopole antennas loaded with a cylindrical homogeneous muscle-equivalent phantom (σ = 0.8 S/m, ε_r = 57, λ_muscle ≈ 90 mm at 434 MHz [25]). The sensitivity of the focus position to phase errors is estimated by assuming two opposite antennas with a phase shift of 5°. The resulting shift in the location of the interference pattern is then 5/360 × 90 ≈ 1 mm. We allow maximum ±1 mm focus-position error and ±5% intensity error in the translation of the planned SAR distribution to the real clinical setting, to be caused by inaccuracies in the amplifier system. In addition, we would like to obtain these accuracies with the lowest calibration intervals. The requirements are summarised in Table I.

Table I.  Design criteria.

Parameter	Value
Number of channels	12
Power per channel	>150 W
Power resolution	<5 W
Relative power accuracy	<±5%
Phase resolution	<1°
Phase accuracy	<±5°
Sampling rate	>5 Hz

System description

Figure 1 provides a schematic overview of the amplifier system designed and shows the relationship between its elements. The path of the signals are visualised clockwise. A Direct Digital Synthesizer (DDS) (Section ‘Direct digital synthesis system’) system generates 12 coherent signals of 10 mW, 433.92 MHz and with individual phase settings. These signals are amplified to a maximum of 200 W per channel by custom-made, but commercially-available amplifiers (Section ‘High-power amplifiers’). Low-loss (<0.1 dB/m) RF-cables (Ecoflex 10 SSB, Germany) are used to direct the high-power signals from the amplifiers to the applicators in the treatment room. Directional couplers (2320/30B, EME HF-Technik, Germany) are placed at the amplifiers to couple a small portion (≈−30 dB) of the amplified signals to newly developed detectors. These detectors measure the gains and phases of both the forward and reflected signals (Section ‘Gain and phase detectors’). A data-acquisition system (Section ‘Data acquisition’) collects the DC detector outputs and performs an analogue-to-digital (AD) conversion. The entire system is mounted in two 19″ racks 2 m in height and is installed in an air-conditioned chamber.

Figure 1. Schematic overview of the designed amplifier system. Fwd is the forward and Rfl is the reflected signal.

The main advantage of this specific design is that we avoid the switching of 12 forward and 12 reflected RF signals, towards a vector voltmeter. Due to the rise-times and lock-times needed for accurate measurements, this switching at the RF level is a slow method. The RF signals are therefore continuously measured by 12 detectors units, that measure (1) the phase of the forward signal and (2) the power of forward and reflected signals. Signal instabilities and inaccuracies often depend on frequency, amplifier gain and temperature of operation. However, we correct for these by a feedback control using measurements after the phase modulation and high-power amplification.

Direct digital synthesis system

The DDS system (Figure 2) is designed around the AD9954 (Analog Devices, USA) and is mounted on seven printed-circuit boards (PCBs). A micro-controller provides both a reference clock signal for the AD9954 and the serial RS-232 communication for modulating frequencies, phases and amplitudes. The phase is modulated with a 14bit resolution. Each AD9954 has an integrated digital-to-analogue converter (DAC) that can generate sinusoidal waveforms up to 200 MHz. The AD9954 is programmed to generate a signal with a frequency of 113.92 MHz at a clock-frequency of 320 MHz. The required frequency with a frequency of 433.92 MHz is obtained by filtering out an alias of the generated signal by a surface acoustic wave (SAW) band-pass filter. Although the output levels are approximately 10 mW, channel 13 is amplified to maximum 0.5 W since it is used as a reference signal for the gain and phase detectors (Section ‘Gain and phase detectors’).

Figure 2. Photograph of the developed DDS system. Channel 13 is amplified and used as a reference signal for the gain and phase measurements.

High-power amplifiers

The DDS signals are amplified using custom-made 270 W AB Class amplifiers (Pavoni Diffusion, Italy). These amplifiers are based on the commercial Alba system described previously in [26], however some changes are made to meet our design criteria. For superficial HT, each amplifier uses its internal 433.92 MHz oscillator. Signals are therefore non-coherent. For head-and-neck HT, the external DDS sources are used instead of the internal oscillator. The output power is remotely controlled by an RS-485 communication protocol at a resolution of ±2 W. The amplifier outputs are protected against high reflected power by circulators which direct this power into a heat-sink. The heat-sinks of both the high-power amplifier and the circulators are protected by 75°C thermal switches (Elmwood Sensors, Thermostat RS 331–556) that allow for automatic shut-down to prevent any damage. Continuous reflected powers up to 50 and 200 W peak values during 1 min, are allowed before automatic shut-down. These shut-downs are accompanied by warnings via a red LED on the front panel and via RS-232 communication.

Gain and phase detectors

Although the amplifiers provide both forward and reflected power measurements, phase information is not provided. Therefore, we developed cost-effective and reliable gain and phase detectors. Each channel has its own detector-unit, which measures the forward gain, reflected gain and forward phase continuously, thereby avoiding RF-switching and waiting for stable signals. The detectors are developed around the AD8302 (Analog Devices, USA), which measures both the gain and absolute phase difference (0 … 180°) between two signals. We combine the information from two of these detectors with a fixed phase-shift between them, to determine the sign of the phase (−180 … 180°). Figure 3(a) shows the photograph of 12 detectors and Figure 3(b) a simplified block diagram.

Figure 3. Photograph of the twelve NLR gain and phase detectors (a). The inputs are RF-forward, reflected and reference and the outputs are DC voltages for gain and phase. A block diagram of the gain and phase detectors is shown in (b).

Data acquisition

The RF signals are continuously measured by the gain and phase detectors. The measurement results, i.e. four DC voltages per channel, are sequentially sampled at 100 Hz per channel, by using a 96-channel analogue-input card (DAQ-2208, Adlink Technology). Calibration coefficients are obtained by linear fits on measured reference data (Figure 4). These coefficients are used for converting the detector voltages into powers and phases according to the following equations:where a₁, …, a₄ are the calibration coefficients.

Figure 4. Example of typical gain and phase detector outputs. Two AD8302 integrated circuits with a phase difference between them are used to determine the sign of the phase.

Calibration coefficient are used to calculate the phases from the voltages measured. Eight coefficients b₁, …, b₈ per channel are obtained from two linear fits on reference data. For example, the phase of the positive slope (↑) of detector V_φ1 is given byBecause the bottom and top of the curves are flat, the linear fits are inaccurate in these areas. To increase the accuracy of the phase measurements, we use the steepest part (at ∼1 V) of the linear fit, while the other measurement is used only to reconstruct the sign of the phase. A more detailed description can be found in [27].

Feedback control

We implemented a proportional-integral-derivative (PID) feedback control to match the power and phases measured to those requested. Since the clinical practice requires stable signals with zero probability for unstable loops, we used a simple feedback with only the proportional term. Although the sampling rate is 100 Hz per channel, the update frequency of the feedback control is only 5 Hz. In addition, the measurements are time-averaged over a period of 1 s for stability.

Graphical User Interfaces

Three graphical user interfaces (GUIs) are programmed in MATLAB (Mathworks, USA) to interface the hardware system. The first GUI is used to perform semi-automatic calibration procedures to obtain the calibration coefficients. The second GUI is used for superficial HT treatments and the third GUI is used for head-and-neck HT (Figure 5). Each GUI provides several common functionalities. First, it is used to load the optimised treatment settings as prescribed by the hyperthermia treatment planning (HTP) tools. Next, these settings are applied to the amplifier system and are feedback controlled. The GUIs monitor the status and warnings of the amplifier system as well. They are used for treatment-security checks, such as antenna reflections and the stability of the feedback control. The GUI's are used to automatically shut-down the amplifiers when high reflected powers are detected (≥20 W averaged over 10 s). Finally, the GUIs are used to visualise and save the clinical treatment data, such as patient information, treatment configuration, pain complaints, pulse-oximeter data and power-time and temperature-time graphs.

Figure 5. Graphical user interface for head-and-neck HT.

Electrical safety

Potential-equalisation connectors are mounted on the DDS system, amplifiers, DAQ, PC and gain and phase detectors, and connected to the hospital potential-equalisation point. A medical separation transformer is used to minimise leakage currents. Finally, the entire amplifier system passed the EMC electrical safety test for medical devices (IEC 601-1, Class I, type B).

Methods: Validation measurements

This section describes the methods used to test the performance and accuracy of the designed amplifier system.

Tests of the amplifier system

The performance of the newly designed amplifier system is determined by using external but highly accurate measurement devices. First, we used an RF spectrum analyser (Agilent E4404B, USA) to measure the spurious emission and harmonic emission, since clean 433.92 MHz signals are required. This RF spectrum is determined for all amplifiers and over the power range of 20–200 W. Next, we established the accuracy of the power and phase measurements. Finally, we characterised the performance of the feedback control.

Accuracy of the power and phase measurements

We used a digital power meter (EMP-442A, Agilent, USA) to measure the accuracy of the power measurements provided by the amplifier system. Since the power meter measures only low powers, we used directional couplers (3020A, Narda, USA) and attenuators (R412720000, Radial, USA). We measured the total insertion loss of this path by using a network analyser (8751A, Agilent, USA). We used 50 Ω terminations for the forward measurements and short terminations for the reflected ones. The power error is defined as the difference between the power readings of the accurate digital power meter and the readings provided by the amplifier system (Figure 6a). We measured the forward power errors in 30 steps between 0 and 200 W. For the reflected power errors we used 30 steps between 0 and 100 W and one step of 200 W. The time between each step was approximately 2.5 s, to provide sufficient settling time for both the amplifiers and the digital power meter. The channels were calibrated sequentially and the total time required for the calibration of both forward and reflected powers of all channels was approximately 30 min.

Figure 6. Configurations to establish the accuracy of the amplifier system by using an accurate power meter (a) or vector voltmeter (VVM) (b).

We measured the phase accuracy by using a configuration that is almost similar to the power one: one DDS channel of the amplifier system provides the reference signal for measurements using a vector voltmeter (Agilent HP8508A). The forward power was set to 30 W. The phase error is defined as the difference between the phase readings of the vector voltmeter and the readings provided by the amplifier system (Figure 6b). We measured the phase errors in steps of 5° between −180° and 180°. The time between each step was approximately 0.5 s, to provide sufficient settling time for the vector voltmeter. The channels were calibrated individually and the total required time for the calibration of all channels was approximately 10 min.

Feedback control

To demonstrate the need for feedback control, we characterised the power and phase changes when the feedback control is disabled.

First, we characterised the phase-shift as a function of amplifier gain. In this measurement we started at 10 W power and we used the phase control algorithm to set the phases of all channels at 0° simultaneously. As a next step, the phase-control was disabled and the powers were increased up to 150 W.

Second, we characterised the signal drifts in both power and phases over a period of 10 min for both low (10 W) and high (100 W) power settings. The phases of all channels were set to 0° simultaneously at the start of the experiment. We use two power levels, since it is expected that the shifts and drifts are temperature dependent and increased temperatures are obtained for high powers.

Finally, we characterised the performance of this feedback control by measuring the step responses in the ranges of 0–150 W and −180° to 180°. For these measurements, we connected all channels to the HYPERcollar in which a muscle-equivalent phantom was inserted. The maximum return loss, i.e. the total of the reflections and cross-coupling was maximum 30% of the forward power.

Results: Validation measurements

RF spectra

We measured the RF spectra of all amplifiers in the power-range of 20–200 W. The spurious emission in a 100 kHz bandwidth around the centre frequency of 433.92 MHz is lower than 60 dBc. The second harmonic frequency of 867.84 MHz is suppressed more than 45 dB and the third harmonic at 1301.8 MHz is lower than 70 dB. Any other frequency is suppressed more than 70 dB.

Measurement accuracy

Figure 7 shows an example of a measurement of the forward power accuracy over the entire power range. From the resulting power errors, we have visualised the mean and standard deviation in Figure 8, which shows the accuracy of both forward and reflected power measurements. The mean error increased from 0% at the initial calibration to maximum 25% at 105 days after the calibration. According to the design criterion of ±5% (Table I), the powers should thus be calibrated at least once per 20 days. The standard deviation of the power errors measured was smaller than 3.5% within the 105-day period.

Figure 7. Example of a measurement of the forward power accuracy. The powers of all channels are measured by both the amplifier system and the accurate power meter (top) and the error between them is computed (bottom).

Figure 8. Forward and reflected power accuracy. The mean and standard errors are the mean and standard deviations of the errors over the entire power range from 0 to 150 W. The black dashed lines indicate the maximum errors according to the design criteria. Note that after the experiment, the powers are re-calibrated with an interval period of 1 month.

Figure 9 shows an example of a measurement of the phase accuracy over the entire phase range. From the resulting phase errors, we have visualised the mean and standard deviation in Figure 10, which shows the accuracy of the phase measurements. The mean phase error increased from 0° at the initial calibration to maximum ±8° at 105 days after calibration. Calibrating the phase once per 2 months is thus sufficient to provide accurate signals as prescribed by the design criterion of ±5°. Within the 105-day period, the standard deviation of the measured power errors is smaller than 2°. Note that although the standard deviation of channel ten is not consistent with the others, it remains within the design criterion.

Figure 9. Example of a measurement of the phase accuracy. The phases of all channels are measured by both the amplifier system and the HP8508A VVM (top) and the error between them is computed (bottom).

Figure 10. Phase accuracy as a function of the time after calibration. The mean and standard errors are the mean and standard deviations of the errors over the entire phase range from −180° to 180°. The black dashed lines indicate the maximum errors according to the design criteria. Note that after the experiment, the phases are re-calibrated with an interval period of 2 months.

RF signal drifts

Due to power-dependent temperature rises of the amplifiers, the powers and phases of the delivered signals drift towards a steady-state value. Power and phase drifts when no feedback control is used, are demonstrated in Figure 11. For low (10 W) output powers, the maximum power drift was 0.05 W/min and the maximum phase drift was 0.7°/min. For high (100 W) output powers, the maximum power drift was 0.3 W/min and the maximum phase drift was 0.8°/min. Note that the measurements are time-averaged over a period of 5 s to enhance the visibility of the graphs.

Figure 11. Drift in powers and phases to demonstrate the need for feedback control. The drifts are measured at both 10 W (a,c) and 100 W (b,d) output power, while the feedback control is disabled.

Figure 12 shows the phase changes as a function of power setting. The phase-shift is maximum 65° over the range of 0–150 W.

Figure 12. Phase as a function of power setting, while the feedback control is disabled.

Feedback control

Figure 13 demonstrates step-responses that were measured to characterise the feedback control. The response of the power feedback loop was measured for all channels at both small (10–20 W) and large (10–100 W) power steps. We use the settling time t_s,5% to indicate the time required for an output to reach and remain within a 5% error band. The maximum t_s,5% for a small (10–20 W) power step is 6 s; for a large (10–100 W) power step it is maximum 11 s. The maximum overshoot (Figure 13a) is 15 W for small power steps. For large power steps there is no overshoot.

Figure 13. Step responses of the power and phase feedback control. The phase responses are measured at both 10 W (c) and 100 W(d) output power.

The response of the phase feedback loop was measured for all channels at phase steps from 0° to 180° at both low (10 W) and high (100 W) continuous-power levels. The settling time t_s,5% at low powers is maximum 9 s; for high powers it is maximum 7 s. For both low- and high-power levels, there is no overshoot in the phase response.

Discussion and conclusions

In this study, we have reported a new 433.92 MHz amplifier system for targeted HT applicators. The system meets our design criteria. Twelve channels are provided with either incoherent or phase-controlled coherent 433.92 MHz signals. The maximum power at the antenna feeding points is 200 W per channel. The power setting is applied with a resolution of 2 W and for the phase it is 0.1°. The channels are sequentially sampled at 100 Hz per channel. The measurement accuracy for the power (±5%) is valid for at least 20 days after calibration and for the phase (±5°) it is valid for at least 2 months. The system is being used for daily-routine HT treatments of both superficial and head-and-neck tumours.

Recent improvements of the amplifier system

One disadvantage of the detectors is the logarithmic gain measurement that requires a stable reference signal. The consequence of this sensitivity to the reference signal is also revealed in Figure 8, which shows that the power errors are continuously increasing since the initial calibration. This is explained by an unexpected decrease in the reference power, which was caused by malfunctioning of the RF amplifiers on the DDS printed-circuit boards. We have now solved this by decreasing the current through these amplifiers. Therefore the mean error of power measurements is now more robust and stable over time.

Switching all forward and reflected RF signals towards a vector voltmeter (HP8508A, Agilent, USA) was a potential solution to measure the power and phases [21–24]. However, these VVMs are no longer sold nor calibrated. Furthermore, the RF switching of 24 channels and the required minimum phase lock-time of ∼0.4 s per channel is too slow for our design criterion of the sampling rate (maximum 0.2 s for all channels). Contrary to the VVM switching practice, we measure all RF signals continuously by using newly developed gain-and-phase detectors. Therefore, switching towards the vector voltmeter is avoided, bringing major advantages in measurement sampling rates. Because the RF signals are measured after the phase and amplitude modulation, any gain or thermal-dependent deviations are corrected by using a feedback control.

Although the measurements are calibrated at the position of the antenna connectors, the signals itself are measured before transportation over the high-power RF cables. As a consequence, any deviations caused by the RF cables are not corrected. However, the normal attenuation of the cables is corrected for in the calibration procedures, and only small changes in the position and bending are made when the cables are connected to the antenna connectors. Although it appears to be favourable to measure the RF signals directly at the antenna connectors, it should be noted that these measured signals must be transported to the detectors by long cables.

In our design criteria, we analytically estimated the required phase accuracy for our HYPERcollar to be 5°, which allows for maximum 1 mm focus position error in muscle tissues (λ_muscle ≈90 mm). For fatty tissues, however, (λ_fat ≈200 mm) the maximum focus position error due to 5° phase errors increases to 3 mm. To further investigate the influence of uncertainties in both power and phases on the targeting, it is recommended to perform a Monte Carlo analysis.

For the signal drifts and tests of the feedback loop, we used the HYPERcollar, approximating the patient by a muscle-equivalent phantom to minimise reflections. This setup allows phase-dependent cross-coupling between the patch antennas. Hence, our drift and feedback loop tests contained also coupled signals that were transferred back to the amplifier and the gain-and-phase detectors. These tests thus represents realistic scenarios that can also be expected in clinical settings.

Although the amplifier system is specifically designed for 433.92 MHz, it might work on other frequencies as well, which might be of interest for similar HT applications that work at frequencies such as 144 and 915 MHz. The latest DDS sources of Analog Devices can generate up to 1 GHz directly, and with our method to filter out an alias we can go even higher. Although the gain and phase detectors are specifically designed for 433.92 MHz, they can be used at up to 2.7 GHz when a phase-shift between the two AD8302 detectors is applied. Although the high-power amplifiers (Pavoni Diffusion, Italy) are narrow band, i.e. 433.92 ± 0.07 MHz, an updated version that works on 144 MHz is being developed.

In conclusion, the presented amplifier system operates according to our design criteria, and is therefore suitable for use in both our superficial and head-and-neck HT applicators. It can also be used for other HT applicators that work on the frequency of 433.92MHz.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.