Minimizing Coupling of Power Supply Noise Between Digital and RF Circuit Blocks in Mixed Signal Systems

Transcription

1 Minimizing Coupling of Power Supply Noise Between Digital and RF Circuit Blocks in Mixed Signal Systems Satyanarayana Telikepalli, Madhavan Swaminathan, David Keezer Department of Electrical & Computer Engineering Georgia Institute of Technology 266 Ferst Dr., Atlanta, GA, USA Abstract Isolation of supply noise between disparate circuit blocks is crucial. When powered by the same voltage supply, the switching noise created at the supply node of the digital devices can couple into the power path of the RF circuitry and cause significant performance degradation. Electromagnetic bandgap (EBG) structures, ferrite beads, and split planes are all commonly used to mitigate this problem, but each have drawbacks which can be detrimental to signal and power integrity. Furthermore, previous works in [1] and [2] have shown that by utilizing a power transmission line (PTL) in place of a power plane, one can significantly reduce the effect of switching noise in high speed digital I/Os by preventing the occurrence of return path discontinuities. The method proposed here extends the concept of the PTL to mitigate the effect of supply noise coupling between a set of digital I/O buffers and an RF low noise amplifier (LNA). In this work, the approach is to place a notch with a bandstop frequency corresponding to center frequency of the LNA in the power supply path of the LNA. Therefore, any frequency content of the switching noise close to operating frequency of the LNA is prevented from entering into its supply node. A board-level test vehicle was built to demonstrate this concept with off-the-shelf components. Through theory, simulation, and lab measurements, is has been shown that utilizing this method can reduce the amount of the switching noise that couples into the output of the LNA by 84%. Keywords simultaneous switching noise; noise isolation; electromagnetic band gap; power transmission line I. INTRODUCTION In complex mixed signal systems, the isolation of supply noise between disparate circuit blocks is very critical. Due to the varying voltage swings of different circuit blocks as well as each circuit s sensitivity to power supply noise, it is crucial to isolate digital and RF devices. Digital devices can have relatively large voltage swings as compared to RF signals. For example, the output voltage swing for Low Voltage Differential Signaling (LVDS) and Positive Emitter Coupled Logic (PECL) devices is 400mV and 800mV, respectively. Integrated RF devices, however, can have very small voltages (< -20dBm or approximately 25mV for a 50Ω system). When powered by the same source voltage, the supply noise created at the power supply node of the digital devices can couple into the RF circuitry. If, for example, the digital devices are powered by a 2.5V supply, the conventional design methodology is to design the power distribution network (PDN) such that the maximum supply noise is within 5 to 10% of the supply voltage. Therefore, the maximum tolerated noise magnitude can be upwards of 125 to 250mV. If even 10% of this noise is coupled into the supply node of a low noise amplifier, then 25mV of noise voltage can be injected into the LNA. Consequently, a large amount of switching noise that is injected into the supply rail by the digital circuitry can couple into sensitive RF devices and cause significant performance degradation in the form of reduced gain and linearity, and can even produce a false signal at the output. A common method for mitigating noise coupling is to use an electromagnetic bandgap (EBG) structure. EBGs are repeating etched metal patterns on the power or ground plane that can be optimized to give a specific frequency response. EBG structures can be very effective in noise isolation, commonly achieving over 30dB of isolation. However, these structures can cause signal degradation, due to the presence of many split planes and via transitions, as discussed in [3] and [4]. Consequently, a method is proposed here that aims to mitigate this problem. The objective of this method is to power both digital and RF devices with a single power source while preventing switching noise from coupling from the digital path into the RF signal path. II. PROPOSED METHOD In this proposed configuration, by adding a in the power supply path of the LNA, it is possible to reduce the amount of noise that is coupled between the two circuit blocks. This configuration is shown in detail in Figure 1. Figure 1. Mixed signal noise isolation Suppose the digital buffers are operating at a certain data rate, f d. The LNA has a desired center frequency of f LNA. If the lower-order harmonics of the simultaneous switching noise (SSN) generated by the buffers happen to fall close to or exactly at f LNA, then the noise will be coupled into the LNA and will be expressed at the output. A notch is designed to have a bandstop frequency at f LNA, and is used as the SSN. With the SSN is connected between the PDN and the LNA circuitry, any power supply noise generated by the digital circuit will be attenuated before coupling into the LNA. The power delivery network used for this circuit is a power transmission line. The power transmission line concept is described in detail in [1] and [2]. If the PTL is placed on the same layer are the signal traces, then the signal network and the PDN will share the same ground reference, such that a /14/$ IEEE Electronic Components & Technology Conference

2 continuous current loop is established and therefore removes return path discontinuity (RPD) effects[1][2]. III. SIMULATIONS Simulations were performed to demonstrate this idea in Agilent Advanced Design Systems (ADS). The LNA used in the simulation is a simple common-gate configuration operating at a center frequency of 2.4GHz with V DD = 2.5V [5]. At the operating frequency, the gain of the amplifier is approximately 11.95dB. A. Case 1: Digital buffers are OFF, RF LNA is ON This is the control case, in which no input signal is applied to the digital buffers and there is a -50dBm signal at the input to the LNA. As expected, there is a -41dBm signal at the desired frequency at the output of the LNA due to the ~10dB gain. Figure 4. Frequency spectrum of LNA output Figure 2. RF LNA used in simulations A 10Ω power transmission line is used to serve as the PDN and four 2.5V CMOS buffers are operating with a 1.2Gbps PRBS-8 input. Figure 3 shows the power spectrum of the noise generated at the V DD node of the buffers. B. Case 2: Digital buffers are ON, RF LNA is OFF; no SSN Figure 5a and Figure 5b show the noise voltage at the supply node of the buffers and the LNA, respectively. In this case, even without an input signal to the LNA, there is an output signal at the desired frequency band, as shown in Figure 5c. This is due to the switching noise from the digital circuits that is coupling into the supply node of the LNA. Figure 3. Spectrum of noise due to SSN Most of the switching noise generated by the digital buffers will be at the switching frequency, 1.2GHz. However, there will also be noise generated at the 1 st harmonic, 2.4GHz. This corresponds to the operating frequency of the LNA, so any coupling between the two devices will cause noise to show up at the output of the LNA. In this simulation, the SSN is a 3 rd order transmission line stub that is optimized to give a stop-band response at 2.4GHz [6] with very high insertion loss. This will prevent noise from the 1 st harmonic of the SSN to couple into the RF output. The circuit is simulated under various conditions, which are summarized below. Identify applicable sponsor/s here. If no sponsors, delete this text box (sponsors). 2288

3 Figure 5. (c) Supply noise on digital size supply noise on RF side (c) LNA output spectrum at 2.4 GHz C. Case 3: Digital buffers are ON, RF LNA is OFF with SSN In this case, the transmission line notch placed in the supply path of the LNA. The has a very high insertion loss at 2.4GHz, and consequently any supply noise generated by the digital circuitry at the bandstop frequency is highly attenuated and does not propagate to the output of the LNA, as shown in Figure 6c. Notice that the peak-to-peak supply noise that is generated by the digital circuits is approximately the same with and without the presence of the notch (Figure 5a and Figure 6a, respectively). Consequently, the only helps provide isolation between the two circuits, and does not contribute more noise. In addition, although the supply noise from other frequencies may couple into the LNA, these signals can generally be ed out further down the receiver chain in a real system. Figure 6. Supply noise on digital size supply noise on RF side (c) LNA output spectrum D. Case 4:Digital buffers are ON, RF LNA is ON; no SSN In this case, both the digital and RF sections are operating simultaneously. Consequently, there is significant noise at the supply node of the LNA and the in-band noise is being coupled into the output of the LNA, as shown by the larger than expected output in Figure 7c. (c) 2289

4 (c) Figure 7. Supply noise on digital size supply noise on RF side (c) LNA output spectrum E. Case 5: Digtial buffers are ON, RF LNA is ON; with SSN When Case 4 is repeated with the presence of the SSN, the noise at the output of the LNA is significantly reduced and the output spectrum of the LNA is nearly identical to that of Case 1, which is desired. Figure 8. Supply noise on digital size supply noise on RF side (c) LNA output spectrum These simulation results show the effectiveness of this strategy. By carefully designing the power delivery network and the SSN, it is possible to isolate the power supply noise between digital and RF circuit blocks in mixed signal systems. In order to further demonstrate this concept, several test vehicles were designed and measured. IV. TEST VEHICLE DESIGN Test vehicles were designed and fabricated to demonstrate the effectiveness of this noise isolation strategy, shown in Figure 9 below. The test vehicles utilize an unterminated 25Ω power transmission line as the PDN. (c) Figure 9. Noise isolation test vehicles with unterminated PTL The low noise amplifier used in all the test vehicles is an off-the-shelf component from Skyworks with a frequency range from GHz and a bias voltage of 2.5V. The operating frequency of the LNA was chosen to be 1.80GHz. Four digital PECL buffers with a supply voltage of 2.5V and 50Ω loads comprise the digital section. The SSN was designed using the transmission line stub matching technique as described in [6] and has a Chebyshev band-stop response with a center frequency of 1.8GHz. The was designed and simulated using a 3D electromagnetic solver, and the physical layout is shown in Figure 10. By collapsing the stubs into a serpentine pattern, the overall area of the was reduced for easier layout onto the PCB. 2290

5 When the -21dBm input is applied to the LNA, and the digital drivers are disabled, the output of the LNA for each test vehicle is shown in Figure 12. One can see that with and without the in the PDN, the output waveforms are very similar, with a peak of approximately -41dBV and -43dBV, respectively. This shows that the presence of the filer does not adversely affect the performance of the LNA. Figure 10. Microstrip line stub layout Figure 11 shows the frequency response of the SSN. Due to imperfections in the PCB material, the resulting bandstop frequency occurs at approximately 1.88 GHz. 0 Insertion Loss (S 12 ) -5 S 12 (db) X: Y: frequency (GHz) Figure 11. Noise isolation test vehicles with unterminated PTL In addition, an identical test vehicle was made in which the transmission line stubs of the were removed, resulting in a straight 25Ω transmission line connecting the supply voltage to the LNA circuitry. Without the stubs, the standard transmission line does not exhibit a bandstop response and does not isolate noise between the digital and RF sections of the board. Other than this change, this test vehicle is identical to Figure 9. V. PTL NOISE ISOLATION MEASUREMENTS The noise isolation test vehicles were tested for the four cases presented previously. For both test vehicles, the input data for the digital buffers was provided by a separate FPGA board which was programmed to provide a 4-bit pseudorandom bit sequence (PRBS-8). An Agilent E8257D signal generator was used to provide a -15dBm input signal for the LNA at 1.88GHz. A 6dB power splitter is used to split the RF signal so that one signal can be used as the trigger source for the oscilloscope to properly view the waveform. Therefore, the amplitude of the signal applied to the input of the LNA is - 21dBm. Figure 12. LNA output without LNA output with SSN Figure 13 shows the frequency spectrum of the output of the LNA when there is no RF input signal but the digital buffers are switching at 610MHz. At this data rate, the switching behavior of the drivers generates a harmonic at 1.83GHz, which is within the bandstop of the SSN. Without the SSN, the peak in the spectrum at 1.83GHz has a magnitude of approximately -64dBV. When the is present, the peak is at -80dBV. Consequently, the presence of the reduces the amount of switching noise that is coupled to the output of the LNA by 16dB. Since the LNA is terminated with a 50Ω load, this translates to an 84% decrease in the noise voltage that is produced at the output terminal of the LNA. 2291

6 Figure 13. LNA output without LNA output with SSN and lab measurements, is has been shown that utilizing this method can reduce the amount of the switching noise that couples into the output of the LNA by 84%. REFERENCES [1] A. Ege Engin and M. Swaminathan, "Power transmission lines: A new interconnect design to eliminate simultaneous switching noise," in Electronic Components and Technology Conference, ECTC th, 2008, pp [2] S. Telikepalli, M. Swaminathan, and D. Keezer, "Minimizing simultaneous switching noise at reduced power with constant-voltage power transmission lines for high-speed signaling," in Quality Electronic Design (ISQED), th International Symposium on, 2013, pp [3] Q. Jie, O. M. Ramahi, and V. Granatstein, "Novel Planar Electromagnetic Bandgap Structures for Mitigation of Switching Noise and EMI Reduction in High-Speed Circuits," Electromagnetic Compatibility, IEEE Transactions on, vol. 49, pp , [4] A. C. Scogna, A. Orlandi, and V. Ricciuti, "Signal and Power Integrity Performances of Striplines in Presence of 2D EBG planes," in Signal Propagation on Interconnects, SPI th IEEE Workshop on, 2008, pp [5] M. Egels, J. Gaubert, P. Pannier, and S. Bourdel, "Design method for fully integrated CMOS RF LNA," Electronics Letters, vol. 40, pp , [6] D. M. Pozar, Microwave Engineering, 4 ed.: Wiley, CONCLUSION The proposed method presented here extends the concept of the power transmission line to mitigate the effect of supply noise coupling between digital and RF circuit blocks. The devices under test are a set of digital buffers and an RF low noise amplifier. If both the digital and RF blocks of a mixed signal design are operating simultaneously and powered by the same source, and a lower-order harmonic of the digital switching noise falls within the operating bandwidth of the LNA, then the switching noise can affect the LNA output. Once present, this noise is very difficult to remove since it is within the operating bandwidth of the LNA. In this work, a method is proposed in which a notch with a bandstop frequency at f LNA is placed in the power supply path of the LNA. Therefore, any frequency content of the switching noise close to f LNA is blocked from entering into the supply node of the LNA. A board-level test vehicle was built to demonstrate this concept with off-the-shelf components, with the digital drivers operating at 610Mbps and the LNA having a center frequency at 1.83GHz. Through theory, simulation, 2292