A FREQUENCY MIXING AND SUB-SAMPLING BASED RF-MEASUREMENT APPARATUS FOR IEEE 1149.4 Juha Hakkinen1, Pekka Syri1, Juha-Veikko Voutilainen2 and Markku Moilanen2 Department of Electrical and Information Engineering, Electronics Laboratory1, Optoelectronics and Measurement Techniques Laboratory2, University of Oulu, Finland Abstract This paper demonstrates the possibility of performing radio frequency (RF) measurements in the IEEE 1149.4 environment using a selected combination of mixing and sub-sampling as the method to down-convert the RF signal into a low frequency suitable for the analogue busses specified by the 1149.4 standard. Signals generated by RF signal generators (2 GHz) and a commercial VCO (3 GHz) were used to demonstrate the operation of the developed RF-to-LF circuitry in cooperation with National’s SCANSTA400 IEEE 1149.4 Analog Test Access Device. The measured repeatability of RF power measurement through the whole system, i.e. from the RF input to the output of SCANSTA400, is ±0.15 dB and the measurement uncertainty of the RF frequency due to measurement instruments and signal sources is ±350 Hz. Interconnect measurements at 2.1 Ghz were also demonstrated. Simple faults such as shorts, opens and missing components may readily be detected at the low-frequency output of the RF-to-LF circuitry. 1. Introduction Reliable, accurate, fast and low-cost testing of products is one of the cornerstones of modern electronics industry. Here the term testing covers production testing, calibration, tuning and maintenance testing. Especially the radio frequency (RF) circuitry presents a tough challenge in the testing of electrical products (e.g. cellular phones, base stations). Testing and tuning of high-frequency parts of electrical systems is extremely time-consuming and requires special, often expensive, test equipment. Furthermore, RF testing will continue to be a major challenge in the future because of new standards (e.g. Bluetooth and IEEE 802.11) and multiband/multi-function products. The emerging testing standard IEEE 1149.4 [1], which extends the well-known digital boundary scan architecture (i.e. JTAG) by adding analogue measurement access to the boundary scan cells, is mainly targeted for low-frequency testing of PCB interconnects and measuring discrete component values, which has previously been demonstrated, for example, in [2] where methods for the calculation of passive component values in some simple situations has been established. However, due to the limited frequency range accommodated by the standard (several decades below RF signals), the standard is not directly applicable to the testing of high-frequency circuitry. However, if the standard could be used for RF testing, the speed (e.g. BIST) and cost (low-frequency low-cost equipment may be used) improvements could be considerable. In addition, this would allow the use of standard (1149.4) test equipment available in the future. In order to allow the measurement of RF signals, some kind of signal processing which extracts the information one is measuring from the RF signal and to present the result as a low-frequency signal (i.e. below 100 kHz) further processed by structures specified in the standard (1149.4) must be used. As the first step towards the application of the IEEE 1149.4 architecture for RF testing, embedded RF test circuitry inside RF ASICs and low-cost measurement instrument are used to replace the conventional and expensive RF testing equipment. Whether this is cost effective when the whole test system is considered depends on integration costs. However, at least RF interfacing and probing problems can be avoided during testing. In the future cost-effective Life-Cycle-Testing (LCT) with high fault coverage and advanced diagnostics may be achieved with totally embedded testing, which provides test reusability during the entire product life cycle. The embedded test architecture may be based on the IEEE 1149.4 architecture with added RF capability, which provides integrated access, stimulus generation and measurement instruments (BIST, Build-In SelfTest). ITC INTERNATIONAL TEST CONFERENCE 0-7803-8580-2/04 $20.00 Copyright 2004 IEEE Paper 19.2 551 fi (t) fc τ t T fo(t) fi(t) fo(t) τ t T Figure 1 The sub-sampling process In developing the embedded RF testing the first step is to investigate the possible frequency down-conversion techniques in cooperation with the 1149.4 low-frequency architecture. The accuracy of the measurement of the translated RF signal property and complexity of the required circuitry are the principal issues when evaluating the usability of different RF test structures for further use, e.g. for ASIC realizations. Therefore, the specific aim of this project was to demonstrate that it is possible to perform simple high-frequency measurements by suitable RF analogue circuit structures (i.e. RF-to-LF signal processing) in cooperation with the IEEE 1149.4 specific structures from target RF frequencies from 2 to 3 GHz. The Apparatus There are several ways to translate a high-frequency signal down to low frequencies, out of which frequency mixing [3] and sub-sampling [4] are the ones most often employed. In the design reported here a combination of these methods was used. Mixing alone would certainly |F(ω)| f ∆f fc 2fc 3fc ∆f ∆f ∆f ∆f ∆f ∆f work for us, but the combination of the two adds freedom in system design. Additionally, sub-sampling from intermediate frequencies may even be directly performed by utilizing the circuit structures specified in the standard as demonstrated in the literature [5]. However, sub- The sub-sampling process can now be straightforwardly presented in the frequency domain using equation (1). As shown in Fig. 2, a copy of the narrowband signal at fRF is produced around multiples of the sampling frequency. Now, a low-pass filter, for example, may be used to extract only the desired harmonic for further processing. 110 1 100 fLO (≈ 1.9 GHz ) φ -3 dB T φ S/H φ fout (to SCANSTA 400) ( ≈ 50 kHz ) 80 70 60 50 fclk=9.985050 MHz fclk=9.988390 MHz fclk=9.991011 MHz 40 2.01 2.02 2.03 2.04 2.05 2.06 2.07 2.08 2.09 f vco/GHz f clk (≈ 9.995 MHz) 552 -2 -3 -5 2 Figure 3 Structure of the RF-to-LF processing unit -1 -4 30 Paper 19.2 fclk=9.985050 MHz fclk=9.988390 MHz fclk=9.991011 MHz 0 90 fout /kHz ≈ 4 GHz band -stop fIF ( ≈ 100 MHz ) (1) n = −∞ n ≠0 ≈ 2 GHz band -pass pulse shaper ∞ ∑ (sin(nπd )) /(nπd )F (ω − nωc ) , 2.2 Realization The structure of the RF-to-LF processing unit is shown in VCOOUT fvco ( ≈ 2 GHz ) G (ω ) = dF (ω ) + d f Figure 2 Sub-sampling in the frequency domain LRPS-2-25J 2.1 Sub-sampling If a signal is sampled by a rate, which is less than the frequency of the signal, the signal is said to be subsampled [4]. In this case, if the signal bandwidth is small, a copy of the original signal is produced at low frequencies by the well-known process of aliasing. The sub-sampling process is shown in Fig. 1 and can be presented mathematically as [4] where G(ω) = Fourier transform of fo(t), d = τ/T, F(ω) = Fourier transform of fi (t), τ = width of the sampling pulse, T = 1/fc = time between samples and ωc = 2π/T. { { fRF =nfc + ∆f { { fc { { |G(ω)| sampling alone is not conveniently done from frequencies above 1 GHz. Therefore, linear mixing was used to shift the input signal to the 100 MHz range, where sampling is more convenient as far as noise aliasing and the speed of the sampler are considered. Due to its familiarity, frequency mixing is not treated theoretically here (e.g. [3]). ∆fvco/kHz 2. Figure 4 The RF-to-LF measurement unit (a) 2.1 2 2.02 2.04 2.06 2.08 2.1 fvco/GHz (b) Figure 5 (a) Measured fout vs. fvco and (b) measurement error 3. Experimental Results 3.1 Characterization The RF-to-LF unit was characterized by measuring the output signal of a laboratory RF signal generator (Marconi Instruments 2024). Agilent 331220A was used as the LO generator and HP 4396A Spectrum analyzer to measure the reference and the LF output. The measured final LF output frequency as the function of the input frequency (fvco) is shown in Fig. 5 (a). Given -31 -32 -33 -34 -35 -36 -37 -38 -39 -40 -41 2.0GHz 2.02GHz 2.04GHz 2.06GHz 2.08GHz 2.1GHz 0 2.01GHz 2.03GHz 2.05GHz 2.07GHz 2.09GHz -10 -20 Pout/dBm Pout /dBm Fig. 3 and the actual in-a-box realization in Fig. 4. To allow reference measurements by traditional RF measuring equipment, the input RF signal (fvco) is first split into two by a power splitter (LRPS-2-25J [6]). The input signal is then converted to an intermediate frequency by a mixer (MBA-15L [7]) and finally subsampled to the proper output frequency by a S/H circuit (constructed around MAX4619 [8]). The recommended frequencies for the different signals in Fig. 3 when fvco ≈ 2.0 GHz are: fLO ≈ 1.9 GHz ⇒ fIF ≈ 100 MHz, fclk ≈ 9.995 MHz ⇒ fout ≈ 50 kHz (i.e. suitable for structures found in National’s SCANSTA400 IEEE 1149.4 Analog Test Access Device [9]). -30 -40 fmix=1.9GHz -50 fmix=1.95GHz -60 fmix=1.99GHz -70 -2 -1 0 1 2 3 -80 0 10000 20000 30000 Pin/dBm 40000 50000 60000 fout/Hz (a) Figure 6 (a) Pout vs. Pin and (b) Pout vs. fout (b) the signals shown in Fig. 3, the output frequency can be calculated using the following equation: fout = (fvco – fLO) – nfclk, where n = 10,11,…20 (2) Equation (2) can also be used to calculate the original RF input frequency from the measured LF output frequency. Thus, the calculated error between the reference measurement and the calculated output frequency is shown in Fig. 5 (b). The error caused by the RF-to-LF unit is in the kHz range, and is directly caused by the inaccuracies of the LO and clock frequencies. In other words, according to the measurements, RF frequency may be accurately measured, after the fLO and fclk are precisely known. RF-to-LF power measurement results are shown in Fig. 6. The high attenuation evident in these figures can y = 70.895x - 25.471, Tuning sensitivity is 70.895 kHz/V 71.4 110 100 71.2 (dfout/dVtune)/kHz/V 90 fout/kHz 80 70 60 50 measured 40 LMS 30 20 71 70.8 70.6 70.4 70.2 10 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 70 0.5 Vtune/V 0.7 0.9 (a) 1.1 1.3 Vtune/V 1.5 1.7 1.9 (a) Output power (output power variation is -36.12 dBm-(-38.1 dBm)=1.98dB). Tuning linearity (non-linearity is +/-0.05543kHz). 0.06 -36 0.05 -36.5 0.03 Pout/dBm fout-ftrend/kHz 0.04 0.02 0.01 0 -37 -37.5 -0.01 -0.02 -38 -0.03 -0.04 0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 Vtune/V Figure 7 (b) (a) VCO frequency tuning characteristic and (b) tuning linearity -38.5 0.5 0.7 0.9 1.1 1.3 Vtune/V 1.5 1.7 1.9 (b) Figure 8 (a) Delta modulation sensitivity and (b) Pout vs. Vtune and output power variation Paper 19.2 553 3.2V 2.7V FAN 2500 Vtune fvco VCO 0.7…1.2V via the modulation index method (see http://www.minicircuits.com/appnote/an95004.pdf) was 450 kHz. 3.23…3.98 GHz VCO box: -3 dB attenuator (LAT-3) CLK box: 74F04 hex inverter Figure 9 VCO and 26 MHz crystal clock (for the sub-sampler) readily be linked to the attenuation caused by the components used in the RF-to-LF signal path. 3.2 VCO measurements After the initial measurements, which provided a calibration database for real RF-to-LF measurements, some VCO specific measurements were made. In these measurements the Marconi Instruments 2024 was used as a VCO by controlling the output frequency with an external analogue voltage. Fig. 7 (a) shows the behaviour of the VCO output frequency as a function of the VCO control voltage (Vtune). The subtraction of the measured data from the best-fit line, i.e. tuning linearity, is shown in Fig. 7 (b) along with the measured tuning non-linearity of ± 0.06 kHz. The delta modulation sensitivity (i.e. dfout/dVtune) is shown in Fig. 8 (a) and the measured behaviour of the RF output power vs. the tuning voltage in Fig. 8 (b). The measured modulation bandwidth, which was measured After the VCO-type measurements made with the Marconi Instruments 2024, test circuits shown in Fig. 9 were built around a real life VCO operating around 3.6 GHz and a 26 MHz crystal oscillator to allow more realistic RF device measurements. Due to the higher input frequency, some of the components (e.g. mixer) in the RF-to-LF unit had to be replaced by higher frequency equivalents. The VCO frequency vs. VCO tuning voltage characteristic, which was measured using the RF-to-LF circuitry, i. e. measurement taken from the LF output, is shown in Fig. 10 (a). The measured tuning sensitivity of 237.8 MHz/V matches the one given in the data sheets of the VCO (250+/-3 MHz/V). The measured tuning linearity is shown in Fig. 10 (b). Fig. 11 (a) gives the attenuation caused by the RF-to-LF circuitry, i.e. the difference between the RF input power and LF output power of the circuitry, as a function of the VCO tuning voltage in three measurement runs. As shown, at a given tuning voltage, the error in power measurement is roughly ±1.5 dB. Similarly, in Fig. 11 (b) the frequency measurement error is shown to be 3.44 -28 -30 3.4 Attenuation/dB fVCO(L)/GHz -32 3.36 y = 0.2378x + 3.1093 3.32 measured 3.28 LMS -34 -36 -38 -40 3.24 -42 3.2 0.4 0.6 0.8 1 -44 1.2 0.6 0.7 0.8 0.9 1 Vtune/V Vt/V (a) (a) 1.1 1.2 1.3 300 26 100 24 ferror/ppm fVCO(L)-fLMS/kHz 200 0 -100 22 20 -200 -300 18 -400 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 Vtune/V 16 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 Vtune/V (b) Figure 10 (a) fVCO(L) (fVCO measured from the LF output) vs. Vtune and (b) tuning linearity Paper 19.2 554 (b) Figure 11 (a) Attenuation caused by the RF-to-LF circuitry and (b) frequency error of the measurement -24 -24.5 Signal generator fin ≈ 3 GHz…3.1 GHz -25 MBA-35L Signal generator -3 dB φ fmix ≈ 2.9 GHz…3 GHz S/H T SCANSTA 400 φ φ Attennuation/dB -25.5 -26 -26.5 -27 average maximum minimum average scansta maximum scansta minimum scansta attennuation of box (Pin=0dBm) -27.5 pulse shaper -28 -28.5 fCLK (26.00001) MHz) -29 Figure 12 Structure of repeatability measurements 3 3.02 3.04 3.06 3.08 3.1 fin/GHz around 26 ppm. Note, however, that the measurements of the real-life VCO are hard to make in this kind of a system due to the inherent instability of the VCO output signal. Random changes in frequency and output power during measurement make accurate measurements hard to perform. Therefore, further analysis of the repeatability of measurements was made using a stable reference signal. 3.3 Repeatability of measurements Since the output signal of a VCO is not very stable, it is hard to estimate the repeatability of measurements when using a VCO as the DUT. Frequency drift and power Attennuation/dB -24 Attenuation of the RF-to-LF circuitry and output power of SCANSTA400 fluctuation of the VCO itself would mask the frequency and power errors of the RF-to-LF conversion, the repeatability of which is of interest here. Therefore, as shown in Fig. 12, in the following measurements an RF generator was used as the signal source once more. Due to the frequency limitations of the signal generators fin was limited to between 3.0 GHz and 3.1 GHz. A spectrum analyzer was used to measure the LF signals after the RF-to-LF device and SCANSTA400 [9]. Attenuation produced by the RF-to-LF conversion and SCANSTA400 and the frequency measurement errors could readily be calculated from the known fout, fclk, fmix and fin and input and output power levels. In repeatability measurements the input frequency was swept ten times from 3 GHz to 3.1 GHz. The attenuation of RF-to-LF conversion and the output power from SCANSTA400 are shown in Fig. 13. -25 -26 -27 -28 -29 3 3.05 fin/GHz 3.1 (a) -56 -56.5 P/dBm Figure 14 -57 -57.5 -58 -58.5 3 3.05 fin/GHz 3.1 (b) Figure 13 (a) Attenuation of RF-to-LF circuitry and (b) output power of SCANSTA400 SCANSTA400 attenuates the incoming LF signal by 30 dB due to the resistivity of the MOS switches used to connect the signal to the internal IEEE 1149.4 analogue busses. This attenuation is removed from the measurements, when necessary, to make the results more comparable. Fig. 14 shows the maximum, minimum and mean attenuation of RF-to-LF circuitry and the maximum, minimum and mean output power of SCANSTA400 (+30 dB), as calculated from the results of the 10 measurements mentioned above. As shown, the attenuation of the RF-to-LF circuitry and the output power of SCANSTA400 vary about ±0.125 dB and ±0.15 dB, respectively. The apparent difference in the slopes of these curves, which can be calibrated out from the actual measurement results, is caused by the different RF input power used during measurements (i.e. needed to overcome the 30 dB attenuation produced by SCANSTA400). The slope of the attenuation curves of the RF-to-LF circuitry was found out to be the same as Paper 19.2 555 4. 5.1 5 4.9 f error/kHz 4.8 4.7 4.6 4.5 4.4 4.3 4.2 2.98 3 3.02 3.04 3.06 3.08 3.1 3.12 fin/GHz Figure 15 Frequency measurement error of the RF-to-LF circuitry the slope of the output power of SCANSTA400, when 0 dBm input was applied in both measurements (see curve labelled as ‘attenuation of box (Pin=0dBm)’ in Fig. 14). According to these observations, it seems that some kind of clipping occurs inside the RF-to-LF circuitry when the input RF-power approaches 0 dBm. The frequency measurement error of the RF-to-LF circuitry is shown figure 15. The frequency measurement error was calculated by equation ferror = fin - fcalc, where fcalc = fout + 4fclk + fmix. The almost constant frequency measurement error produced by frequency offsets of the signal generators ranges from 4.3 kHz to 5kHz. According to the figure, the measurement uncertainty is roughly ±350 Hz. Interconnect measurements At this point it was recognized, that the measurements so far did not quite fulfill the basic premises of the IEEE 1149.4 standard – namely interconnect measurements. The low-frequency interconnect measurements proposed by the standard are for the most part meant for detecting faulty opens and shorts on the printed circuit assembly (PCA), which arise from faulty soldering and problems in the manufacturing process, for example. However, these low-frequency measurements cannot be used for detecting faulty short and opens in RF interconnects, where DC blocking (i.e. series capacitors) is often used. Additionally, the standard does not lend itself at all to the measurement of the actual RF properties of the interconnect transmission lines and RF components, i.e. characteristic impedances and insertion loss, for example. In the following paragraphs, the RF-to-LF circuitry will be used to demonstrate the detection of faulty shorts and opens in an RF transmission line operating at 2.1 GHz. There are several ways to detect a faulty open or short at the end of an RF transmission line. All the basic methods investigated here are based on the detection amplitude and/or phase of the signal reflected back from the load (i.e. open or short) as shown in Fig. 16. Simple amplitude detection is especially useful, when the intended, nonfaulty, load provides a good impedance match, i.e. the difference between the magnitude of the reflected wave between the non-faulty (i.e. ideally a zero reflected wave Z0 Z0 Z0 Z0 (a) (b) ∆φ Z0 Z0 Z0 Z0 (c) Figure 16 Paper 19.2 556 (d) Different reflection cases. (a) Perfectly matched load, (b) faulty open, (c) faulty short and (d) faulty open, with nonmatched load. RF source 50 Ω TL Coupler 0.05 0.04 LO source RF LF G + LP STA400 LF-out/V lock open short 0.03 Reflected signal CLK source 0.02 0.01 50Ω 0 -0.01 -0.02 -0.03 -0.04 REF source -0.02 (a) -0.01 0 0.01 0.02 0.03 0.02 0.03 time/s (a) 0.05 0.04 short long LF-out/V 0.03 0.02 0.01 0 -0.01 -0.02 -0.03 -0.04 -0.02 -0.01 0 (b) Figure 17 Measurement setup for the reflection measurements. (a) Schematic and (b) laboratory setup. as in Fig. 16. (a)) and faulty (ideally the reflected wave has the same magnitude as the input wave Fig. 16. (b) and (c)) cases is large. However, in situations where the original load is poorly matched to the transmission line impedance, i.e. the reflected wave does not change its magnitude whether there is a fault or not, the only way to detect the fault is through measurements of changes in the electrical length of the transmission line, which may be accomplished by measuring the phase difference ∆φ of the reflected waves as in Fig. (d), for example. This simple measurement may, however, miss the fault if ∆φ happens to equal 360° due to the different complex reflection coefficients of the load and the fault or the exact position of the fault. On the other hand, the possibility of a miss is relatively small. All reflection measurements were made using the measurement setup of Fig. 17. The 2.1 GHz test signal is connected to the test transmission line through a directional coupler (Coupler), which connects only the reflected wave to the RF-to-LF apparatus. Since phase measurements can only be made relative to some phase 0.01 time/s (b) Figure 18 Signals at the output of the LF amplifier/low-pass filter, when (a) the termination of a 50 Ω transmission line is 50 Ω, ∞ Ω (open) and 0 Ω (short) and (b) when the length of an open-ended transmission line is changed by approximately ¼ of the wavelength. reference, all signal sources (RF, LO, CLK and REF) are frequency locked, out of which REF source provides the reference signal against which the 100 Hz output of the RF-to-LF circuitry is compared using an oscilloscope. Due to the high level of attenuation caused by the RF-toLF circuitry and the SCANSTA 400 PCB (STA400), an LF amplifier low-pass-filter combination (G + LP) was used to boost the LF signal of interest. Furthermore, the low-pass filter reduces the high-frequency components of the RF-to-LF circuitry and thus makes the output more readable from the screen of the oscilloscope. As shown in Fig. 18. (a) the low-frequency output may be used to detect differences in the magnitude of the reflected wave in case of a good and poor matches. Additionally, the phase of the output waveform clearly indicates the type of load, i.e. open or short. Since it is easy make out the difference between the amplitudes of the well matched 50 Ω load case and the two cases Paper 19.2 557 mimicking most common failure mechanisms (i.e. electrical short and missing or poorly soldered component at the end of the transmission line), this test seems quite practical and useful. In Fig. 18. (b) an open-ended transmission line was first measured and the output signal then compared to that of a shorter transmission line. The different electrical length clearly manifests itself in the LF output of the RF-to-LF circuitry. In this case, the difference between the lengths of the transmission lines was roughly ¼ of the wavelength, thus producing approximately 90° phase difference. The signals in Fig. 18. (b) are quite noisy due to the harmonics of the sampling frequency, which would suggest that the detection of small differences in transmission line lengths would be hard to detect. However, the signals could very easily be filtered to those superimposed on the measurement results with a higher order low-pass filter than used during measurements. Even so, the detection of phase differences is by far more difficult than the detection of amplitude differences. However, unmatched terminations with reflection coefficient magnitudes of one are quite rare in practice, thus, the amplitude detection may often be used in conjunction with the phase measurement for increased reliability of fault detection. 5. Discussion The use of sub-sampling causes many problems in the RF-to-LF circuitry. The input RF signal must be mixed to quite a low frequency (100 MHz) to allow the use of a typical not-so-fast switch, which requires a very high LO frequency. With a more advanced high-frequency switch, the mixer could be used for coarse frequency tuning of the input signal, or it would not be needed at all. In any case, sub-sampling has its problems, similar to those that have prevented its wide use in any radio receiver (speed and noise, for example). Positioning the LF output signal down to the 0-100 kHz band is also problematic. The correct mixing frequency can only be calculated if we knew the frequency of the input signal beforehand. Otherwise the frequency band of interest must be swept by tuning the mixing frequency. Searching the signal by manual tuning is quite laborious, wherefore an automatic sweep-and-measure system will be needed in practical applications. This may slow down the entire measurement to an unacceptable extent. Overall, the testing time using the described circuitry is at the moment comparable to that of using conventional dedicated RF test instruments. In other words, attaching measurement cables (or probes) and producing/reading the measurement result is roughly the same. The savings in testing time come into play only after a measurement Paper 19.2 558 circuit similar to the one proposed here is embedded into the DUT itself. In this case, the test access is always available without any special test arrangement or ATE, for example. Additional savings in time can be achieved through optimized test software, which may be partially implemented in hardware for further operational speed. Therefore, the integration of the RF-to-LF circuitry as part of the equipment under test is the road to test cost savings during production testing and even more so in field tests, where the production ATE and test equipment may not be readily available. However, even if the presented circuit in its current form does not provide any savings in testing time, the cost savings compared to conventional pieces of equipment is considerable. The total cost of the described circuit is in the $100-or-less range, which is from one upto two decades below that of the combined cost of a commercial RF power and frequency meter. Naturally, savings due to the simplicity of the circuit come at the expense of reduced accuracy. Attenuation of the RF-to-LF circuitry (≈ 26 dB) and SCANSTA400 (≈ 30 dB) cause the signal level to drop very low (i.e. close to the noise level of the measurement instruments). For example, during the VCO measurements the attenuation had to be compensated for by an LF-amplifier, which was connected between the RF-to-LF circuitry and SCANSTA400. To overcome the attenuation of the RF-to-LF circuitry the sampling circuit should be redesigned for more broadband operation. If the sampler were fast enough and, thus, there were no need for a mixer, the attenuation would certainly drop to a more manageable level. Accurate VCO measurements are hard to make due to the inherent instability of any unlocked VCO. Several measurements are needed to average results to a meaningful accuracy, which makes measurements slow to perform. Interconnect measurements at RF may be performed using the proposed apparatus. Simple faults, such as shorts, opens and missing components may readily be detected at the low-frequency output of the RF-to-LF circuitry. Here, only simple reflection measurements at a constant frequency were briefly investigated. Interconnect faults, which change the magnitude of the reflected wave are easily detected, since they don’t require a phase reference. However, if some kind of low-frequency reference signal is available, faults creating phase changes in the reflected wave, e.g. shorts, opens or wrong termination impedances, may be detected through phase measurements. In our case, the phase reference was generated by an auxiliary low-frequency generator and by carefully frequency locking all signal generators, which may not be possible in a real-life implementation. Due to the many promising results, the next goal of the ongoing project is to design a selection of RF-to-LF analogue circuit structures needed to produce IEEE 1149.4 compatible RF Analog Boundary Modules (RFABMs) for basic RF measurements and to realize them as an integrated circuit. 6. Conclusions It seems, according to the measurements, that it is possible to make simple RF measurements using a combination of mixing and sub-sampling as the method to down-convert the RF signal into a low frequency suitable for the analogue busses specified in the 1149.4 standard. However, for accurate measurements the RF-toLF unit has to be carefully characterized and the gathered information thoroughly used to de-embed the unit from the measurement system. The repeatability of RF power measurement through the whole system, i.e. from the RF input to the output of SCANSTA400, is ±0.15 dB. The measurement uncertainty of RF frequency caused by measurement instruments and signal sources is ±350 Hz. Interconnect measurements at RF may also be performed using the proposed apparatus. Simple faults, such as shorts, opens and missing components may readily be detected at the low-frequency output of the RF-to-LF circuitry. 7. [1] [3] [4] [5] [6] [7] [8] Acknowledgements This work was carried out in the project “Development of the testability of Mixed-Signal circuits” funded by the Technology Development Center of Finland (TEKES), Nokia and Elektrobit Group. 8. [2] [9] T. Saikkonen, J-V. Voutilainen, M. Moilanen, “Some Methods to Calculate the Values of Passive Components from the Measurements Made with an 1149.4 Compliant Device”, 2nd IEEE International Board Test Workshop, October 2-3, 2003, Charlotte USA, 10p H. L. Krauss, C. W. Bostian, F. H. Raab, ”Solid State Radio Engineering”, John Wiley & Sons, Inc, USA, 1980 R. Mason, S. Ma, “Analog DFT using an undersampling technique”, IEEE Design & Test of Computers, Oct-Dec 1999, pp. 84-88. S. Sunter, K. Filliter, J. Woo, P. McHugh, “A General Purpose 1149.4 IC with HF Analog Test Capabilities”, Proceedings IEEE International Test Conference, 2001, pp. 38-45 Mini-Circuits, “LRPS-2-25J Power Splitter”, http:// www.minicircuits.com/cgi-bin/spec?cat=splitter& model=LRPS-2-25J&pix=qqq569.gif&bv=4, MiniCircuit’s online data sheets, May 2003 Mini-Circuits, “MBA-15L Frequency Mixer”, http://www.minicircuits.com/cgibin/spec?cat=mixer &model=MBA15L&pix=sm2bc.gif&bv=4, Mini-Circuit’s online datasheets, May 2003 Maxim, “MAX4617, MAX4618, MAX4619 HighSpeed, Low-Voltage, CMOS Analog Multiplexers/ Switches“,http://www.maxim-ic.com/quick_view2. cfm/qv_pk/2064, Maxim’s online data sheets, May 2003 “SCANSTA400 IEEE 1149.4 Analog Test Access Device”, National Semiconductor Advanced Information, August 2000 References IEEE Std 1149.4-1999, “Standard for a Mixed Signal Test Bus”, IEEE, USA, 2000 Paper 19.2 559