| Both sides previous revisionPrevious revisionNext revision | Previous revision |
| university:tools:pluto:users:amp [14 May 2018 22:46] – [Distortion] Robin Getz | university:tools:pluto:users:amp [21 Jan 2019 14:13] (current) – [Peak to Average] Robin Getz |
|---|
| ^ Power in mW | Power in dBm | | ^ Power in mW | Power in dBm | |
| | 0.1 mW | -10 dBm | | | 0.1 mW | -10 dBm | |
| | | 0.3 mW | -5 dBm | |
| | 1 mW | 0 dBm | | | 1 mW | 0 dBm | |
| | | 3.2 mW | 5 dBm | |
| | 10 mW | 10 dBm | | | 10 mW | 10 dBm | |
| | | 32 mW | 15 dBm | |
| | 100 mW | 20 dBm | | | 100 mW | 20 dBm | |
| | | 316 mW | 25 dBm | |
| |
| A doubling of output power (from 1mW to 2mW) is only +3dBm. A gain of +20dBm, is output power increasing by a factor of 100 times in mW. | A doubling of output power (from 1mW to 2mW) is only +3dBm. A gain of +20dBm, is output power increasing by a factor of 100 times in mW. |
| |
| whether expressed in percent in dB, PAPR is dimensionless quantity. | whether expressed in percent in dB, PAPR is dimensionless quantity. |
| | |
| | When dealing with signals and amplifiers, it is the peak that we need to be concerned about, not the average power in the signal. Different types of modulation schemes have different peak to average power, and this needs to be taken into account. |
| |
| |
| | 15 | 31.6 | 1.257 V | 3.556 V | | | 15 | 31.6 | 1.257 V | 3.556 V | |
| | 20 | 100 | 2.236 V | 6.324 V | | | 20 | 100 | 2.236 V | 6.324 V | |
| | | 25 | 316 | 3.976 v | 11.246 V | |
| | |
| The question is, how do we get +20dBm (6.324V <sub>peak-peak</sub>) out of a system, when the power supply is limited to 5V? The trick is in how we connect the output stage. The output stage (RFOUT) is connected to Vcc through the inductor L1. From a DC perspective, inductors become short circuits, and RFOUT is setting at 5.0V, allowing a 10V<sub>peak-peak</sub> swing from the amplifier. This is also why it is AC-coupled by the output capacitor before it attaches to the antenna. | The question is, how do we get +20dBm (6.324V <sub>peak-peak</sub>) out of a system, when the power supply is limited to 5V? The trick is in how we connect the output stage. The output stage (RFOUT) is connected to Vcc through the inductor L1. From a DC perspective, inductors become short circuits, and RFOUT is setting at 5.0V, allowing a 10V<sub>peak-peak</sub> swing from the amplifier. This is also why it is AC-coupled by the output capacitor before it attaches to the antenna. |
| |
| |
| ===== Measurements ===== | ===== Measurements ===== |
| | |
| | ==== Signal To Noise Ratio ==== |
| | |
| | When we consider the performance of a communication link, in the most basic sense, we are interested in the bandwidth and power of the transmitted signal. Bandwidth is measured from the power spectral density of signal and is also proportional to the bit rate. We define average energy per bit as: |
| | |
| | <m>overline{E}_b = {overline{E}_s}/{log_2(M)}</m> |
| | |
| | where <m>M</m> is the order of the modulation scheme and <m>overline{E}_s</m> is the average symbol energy. Note that <m>overline{E}_b</m> is of units Joules/bit. When considering an Additive White Gaussian Noise (AWGN) channel, we can write the power efficiency as <m>{overline{E}_b}/{N_0}</m>, where <m>N_0</m> is the noise power spectral density (PSD). Since AWGN is flat in frequency, or equal across all frequencies, channel noise can typically be measured over 1 Hz of spectrum. Therefore, <m>N_0</m> is in units Watts/Hz <m>right</m> Joules. <m>{overline{E}_b}/{N_0}</m> can be alternatively understood as normalized signal-to-noise ratio (SNR) and is related by the bitrate <m>R_b</m> and channel bandwidth <m>B</m> |
| | |
| | <m>SNR = {overline{E}_b}/{N_0}*{R_s}/{B}.</m> |
| | |
| | SNR is commonly expressed in SNR in decibels (dB) and the equation above can be rewritten as: |
| | |
| | <m>SNR_{dB} = 10log_{10}({overline{E}_b}/{N_0}) + 10log_{10}({R_s}/{B})</m> |
| | |
| | When determining SNR of a signal it is important to understand that signals are band limited unlike noise. Therefore, as the observation bandwidth of the signal is increased the SNR becomes worse. |
| | |
| | ==== Noise Generation and Power ==== |
| | |
| | Calculation of power is a non-trivial exercise in practice and in many situations loosely defined. This is especially true when determining SNR. Let us first consider a unique example where we have scaled exponentials (<m>s(t) = alpha exp(j pi f_c t)</m>) in AWGN. A resulting FFT provided in the figure below, generated from some simple code for two different source signals: |
| | |
| | <code> |
| | r = signal+noise; |
| | % View averaged spectrum |
| | freq = linspace(-bandwidth/2,bandwidth/2,fftLen); |
| | R = reshape(r,fftLen,frames); |
| | R = fftshift(fft(R)); |
| | R_mean = mean(abs(R),2)/fftLen; |
| | plot(freq,10*log10(R_mean)); |
| | </code> |
| | |
| | {{ :university:tools:pluto:users:snr_guess_dual-eps-converted-to-1.png?600 |}} |
| | |
| | |
| | An obvious question to ask would be "What is the SNR of these signals?". However, we cannot directly derive the SNR from this plot alone. If we are using a spectrum or vector analyzer we need knowledge of the Resolution Bandwidth (RBW), which is a function of the FFT bin count and observation bandwidth. Since we already know the observation bandwidth (1 MHz), if we used a FFT of size 1024, then the true SNR is ~10 dB. An important aspect to note from the figure above is the shape of the signals <m>r_1</m> and <m>r_2</m>. In generation these two signals only differ in <m>f_c</m>, although they appear quite different in the figure. <m>r_2</m> was strictly chosen to be within a FFT bin and <m>r_2</m> strattles bins. This difference is due to the window effect of the FFT applied to the signals. This is commonly known as //scalloping// and will be dependent on the window chosen. fred harris (yes that capitalization is correct) provides a detailed overview of different windows and how to compensate for their effects in [harris1978]. When performing digital signal processing it is alway important to understand effects of discrete computations over the infinite resolution of our written equations. |
| | |
| | Now let us connect this to the theoretical concepts. We know that the power spectral density (PSD) is obtained through a Fourier Transform of a signal's autocorrelation function. Formally written as: |
| | |
| | <m>S_{XX}(f) = int{-infty}{infty}{R_{XX}(tau)e^{-j2 pi f tau} d\tau}</m> |
| | |
| | where the autocorrelation of our process or signal <m>X(t)</m> is: |
| | |
| | <m>R_{XX}(t_1,t_2) = E[X(t_1)X^{ast}(t_2)]</m>. |
| | |
| | In the case of AWGN this autocorrelation is simply: |
| | |
| | <m>R_{XX}(t_1,t_2) = sigma^2</m>(if <m>t_1=t_2</m>, <m>0</m> otherwise) |
| | |
| | where <m>sigma^2</m> is the variance of the noise. Therefore, <m>S_{XX}=sigma^2</m> for all frequencies within the observation bandwidth. Therefore, the power of the AWGN signal is simply <m>P_N = sigma^2*B</m>. This result provides a mechanism for determining power from the PSD or frequency domain of a signal. However, it can be useful to calculate signal power from the time domain, and from Parseval's theorem we know power is identical between domains. For an AWGN signal calculating the variance directly provides <m>P_N</m>, but this is not true for non-zero mean processes. In that case, power can be simply calculated based on the squared RMS or squared mean of the process samples. However, a significant amount of data should be collected to get a representative power value of the signal. In MATLAB for a complex input signal the power is simply: |
| | |
| | <code> |
| | signalpower = sqrt(mean(signal.*conj(signal)))^2; |
| | </code> |
| | |
| | However, the power unit is a bit tricker to determine. An instrument will provide the I/Q data in voltage based on some input impedance. Therefore, we can calculate the power directly in dBm easily: |
| | |
| | |
| | <m>P_{dBm} = 10log_{10} ({P_{RMS}}/{1mW}) = 10log_{10} ({{I^2+Q^2}/2}/{R*1mW})</m> |
| | |
| | |
| | where <m>R</m> is the input resistance of the device. For an SDR I/Q values will be based on some uncalibrated ADC values and we typically define them of unit dBFS. Resulting in a power calculation that is not relatable to Watts. If it was desired to use an SDR an a instrument we would need to relate each output of the ADC to a specific input voltage at the antenna. This mapping would allow us to accurate measure quantities in SI units instead of relative units. |
| | |
| | ==== S Parameters ==== |
| | |
| | This data was taken on a [[https://www.keysight.com/en/pdx-x202208-pn-E5080A/ena-vector-network-analyzer?|Keysight ENA E5080A]]: |
| | |
| | First we calibrate things with a cable, and connector, to make sure we see what is happening. We expect this to be a flat line, with 0dB of gain. (it is a cable after all). |
| | |
| | {{:university:tools:pluto:users:cal_s21.png?600|E5080A Calibration}} |
| | |
| | Then we can look at the S12 of the amplifier board. Here we can see gain between 2 and 3 GHz, with the flat part being between 2.4 and 2.5 GHz, just like we hope. |
| | |
| | {{:university:tools:pluto:users:c419_s21.png?600|}} |
| | |
| | {{:university:tools:pluto:users:c419_s21_zoom.png?600|}} |
| | |
| | If we vary the amplitude at a constant frequency, we can see the P1dB point at +5dBm. In order to keep things operating in the linear region, we should make sure not to drive the amplifer board with more than +5dBm. |
| | |
| | {{:university:tools:pluto:users:c419_p1db.png?600|}} |
| | |
| | ==== Results==== |
| | |
| | The yellow line is an antenna, the red line is with the same antenna and the amplifier. You can see the +20dB of transmission at 2.4GHz. |
| | |
| | {{:university:tools:pluto:users:c419_s21_ant_amp.png?600|}} |
| | |
| | |
| | |
| |
