Accorgimenti Progettuali HD & FW CNR Firenze (3 luglio 2009) Emi Receivers & Impulses Receiver Typology Spectrum Analyzer Frequency domain • Usually no pre-selector • Analog • Digital Time domain • No pre-selector • Only Digital (FFT) Receivers (EMI) • Frequency domain Always pre-selector • Analog • Digital • Time domain ?? Frequency Domain Either analog or digital performs a scan tuning frequency by frequency. Pre-selector can be as narrow as the used RBW. Detector can be updated continuously virtually with no time limit (Zero Span) Frequency Domain Structure 10 0 Time Domain Based on Fourier Transform (DFT,FFT) Evaluates N frequency in a single shot T=(Bin/Fs) • If Real input N=(Fs/2)/bin • If Complex input N=(Fs)/bin Pre-selection cannot be narrow otherwise FFTs are distorted CISPR Requirements Quasi Peak Detector over long time (up to 15s Cispr 16-2-3) Filter BW Mask mandatory Overload Factor EMI TD Analyzers Structure I 1 GHz bandwidth at once Because of its nature, no pre-selection is present Full Input power is sent to ADC Basically the only difference from Digital frequency domain is the pre-selector. EMI TD Analyzers Structure II A more realistic one has been later proposed Sub scans (125MHz wide) are sequentially tuned by a down converter. Full Power, however, still enters the mixer EMI TD Analyzers Hurdles Two main difficulties have to be dealt with: • Processing complexity circuitry. It is discussed it is neither a mere single FFT, nor a sheer post processing task. • Dynamic range. It is the real hurdle which prevents using broad range. Windowing (Filter Shaping) Scalloping, Picket Fence Effect DFT (FFT) cannot be used without windowing. Windowing reduces spectral leakage, the “Picket Fence effect” or “Scalloping”. Windowing means Filter shaping. CISPR 16 BW MASK Although somewhere it is stated: • “It is evident that the requirements can be met by applying a Gaussian window function and the STFFT.” It is not straightforward. Very complex Windowing is required to fulfill the mask. Impulse Ideal impulse has no duration. • Impulse duration seen by ADC is only made (shaped) by the preceding Filter. • As a rule of thumb, duration is roughly 2/BW. In the example . • BW = 1GHz (125MHz). Impulse response ~2ns (~16ns) • F Sampling = 2GHz (250MHz) Sampling time = 500ps (4ns) Not missing a single sample is PARAMOUNT Windowing Impulses Impulse can be out of central window. • Because of multiplication by a nearly 0 coefficient, it is practically missed altogether. • The time span where error is acceptable can be very narrow. FFT over overlapped samples. • The only solution is to have FFT on overlapped samples. • This means that more FFTs are computed at same time (parallel) for almost the same samples and then mixed out (filter). Real Time CISPR 16-1-1 specifies Quasi-Peak down to 1Hz repetition rate. • The only possibility to have the correct result is to get every single impulse and all signals in between. • This can be achieved only when processing in Real Time. Things to be done at average sample time: • Sample Windowing . • Calculate all FFTs (not just one, but ~16) required not to miss any single impulse. • Filter out all FFTs done. • Convert all data from Cartesian into Polar coordinate. • Make ALL detectors for ALL frequencies (Bin) calculated (some k). Real Time processing Data is processed as soon as it is sampled Extremely fast calculation needed. All processing has to be done immediately. • Windowing • FFTing (more FFTs have to be made on the same samples) • Peak, QPeak, Rms, Avg, C-Rms, C-Avg. Detectors for ALL frequencies at once (some k!) Impulse (again) Impulse energy is spread out over the spectrum. • The distribution of the spectrum depends on Impulse duration (TP). • A flat spectrum envelope up to 1 GHz requires pulse duration of ~400ps. • Impulse energy is normally expressed in Amplitude per Bandwidth [ (dBµ) V / MHz] Wider BW gets higher power. • The relationship is A1 /A2 =Bw1 /Bw2. Therefore dB(BW1_BW2) =20 log(Bw1 /Bw2). • Hence, the importance of pre-selector: the higher the ratio, between first IF BW and last BW, the bigger the drop in amplitude. • For instance, an impulse of 100dBµV/MHz yields to: 140dBµV on a 100MHz Bw 82dBµV on a 120kHz Bw EMI TD Analyzers Amplitude As shown on the above 2 structures, the ADC is fed by 1GHz Bw and 125MHz Bw. Thus, the loss of dynamic range will be respectively, for a Bw of 120kHz (Cispr 16 band C&D). • 78dB and 60dB! This loss cannot be recovered anymore; just lost! The Hard Task CISPR 16-1-1 specifies Quasi-Peak at different repetition rates (Bands C & D). • 100Hz is the reference (0) which is in turn ~12dB below Peak indication. • At 1Hz spec is -28.5dB below the reference which means: • (12+28.5 ) >40dB below Peak value. In order to work noise floor should be an additional 15-20dB below the minimum specified level to make correct weighting. Some more headroom is needed to have attenuation intervention as well as flatness compensation: say, at best, another 5 dB. As a result, a minimum of 60 to 65 dB are required. All EMI receivers must have at least these figures. The Impossible Task To this dynamic range we have to add the loss due to the above discussed ratio (Pulse Bw / IF Bw) For a 1GHz IF Bw we add 78 dB. This makes the awful dynamic range of 143dB! For a 125MHz IF Bw we add 60dB. This still makes a remarkable dynamic range of 125dB ! All the figures are conservative as real hardware is likely to be worse Impulse (again) From a different point of view Cispr16-1-1 specifies the impulse area as 0.022µVs. A flat spectrum envelope up to 1GHz requires a pulse having 400ps duration. Thus, amplitude is 22e-9/400e-12=55V. Cispr16-1-1 specifies the response at 100Hz to be 60dBµV (1mV). Level that the Rms detector would get measuring a 60dBµV sinusoidal signal. Cispr16-1-1 specifies at 1Hz Qpeak indication will be 28.5 dB below the reference (100Hz). Therefore indication will be 31.5dBµV (~37µV). Hence, the required dynamic range is: 55V / 37.µV= 55/37e-6 =1.75e6Æ ~123dB (plus headroom for weighting and compensation) Circuitry complexity As seen all detectors, but QuasiPeak most, need a real time, no sample lost, continuous processing. Some unsuitable solution, such as saving only bursts, are proposed , but they are just “tricks” and not compliant. • As above seen, all single samples must be kept. It is often argued there is no limitation whatsoever while in reality complex signals cannot be handled. The Main Point: Dynamic Range Hundreds MHz sampling ADC with such a dynamic range does not exist. A “nearly performing” ADC as above does not exist, either. To overcome this difficulty it has been proposed – taking it up again- the “Floating point ADC”. But, does it work under any circumstances? Floating Point ADC Floating point ADC is based on multiple scaled ADCs. It is not a new idea. • According to the level, the appropriate one is chosen. • In the example the increase in dynamic is 30dB. • It behaves like a variable attenuator but it does not need to re-measures as data is already done. Floating Point ADC Floating Point ADC When more signals (much) different in amplitude are present in same ADC band, the selected ADC will be the one which can handle the highest signal. As a consequence, weaker signals are neglected. The result is that the spectrum is either distorted or, much worse, lacking in signals. PMM9010 Structure CISPR16-1-1 Full Compliant 9010 Functional Block Diagram PMM 9030 / 9060 Structure CISPR16-1-1 Full Compliant Conclusions EMI receivers have different approach from Spectrum analyzers due to the fact they have to deal with non repetitive signals as well as impulses. TD Receivers could be seen as pre-compliant EMI receivers - assuming the circuitry is sufficient depending on the one-shot-band they deal with: • • • • The shorter the taken time, the wider the band. The wider the band, the less it complies. The more the compliance, the narrower the band. The narrower the band, the longer the time is taken. Conclusions Pre-selection is not just a “cut off unwanted signals” It is an indispensable part of an EMI receiver, though. Expedients, like “Floating ADC”, although suitable for continuous signals, may not work when measuring impulses. As long as Cispr PULSES, especially low repetition rate, a FULL compliance (Band C&D) , one shot, still remains an impossible task. Accorgimenti Progettuali HD & FW CNR Firenze (3 luglio 2009) Grazie per l’attenzione