Fullscreen HTML5Med Res (??MB)HTML4

Circuit Surgery Regular clinic by Ian Bell Measuring the frequency response of a circuit or device using with a PC sound card, part 1 52 Frequency response concepts This expression, Y(f) = H(f)X(f), implies Input amplitude linearity. That is, at a given frequency, the value of H is fixed – it varies with f, but not with other things such as time or any other characteristic of the input signal, such as its amplitude. H not changing with X (ie, only with f) means that for a particular frequency (f1, say) we could write X=Hf1Y and a graph of X (input) against Y (output) at that frequency would be a straight line (hence ‘linear’). Linearity in combination with H not varying with time means we have what is called a linear time-invariant (LTI) system. A lot of theory behind electronics design is based on the assumption that circuits like amplifiers and filters are LTI systems. Given that we are interested in characterising a circuit’s behaviour with frequency, it makes sense to analyse or measure circuits by applying signals containing a single frequency. We can repeat this for as many frequencies as we need to build a picture of the frequency response. This is not the only way to obtain a frequency response, but it is the most straightforward to think about it. Sinusoidal waveforms are the only waveforms that contain just a single frequency – other wave shapes have content at multiple frequencies. This is the basis of Fourier theory, which states that any waveform can be decomposed into purely sinusoidal components. So, we can obtain a frequency response (in theory, simulation and practical Negative output phase shift (lagging) Positive output phase shift (leading) Output amplitude Amplitude processing (DSP), which has covered the various parts of a generic DSP system and the basics of their theory of operation. In the most recent few articles, we described the implementation of a basic digital filter using a microcontroller. In the last issue, we showed oscilloscope traces for a couple of input and output signals to confirm a low-pass filter example did attenuate higher frequencies. However, what we really want to do to evaluate a filter is to plot its measured frequency response. This month, we will start to look at frequency response measurement in general, not just for the DSP example. The frequency response of a circuit is the variation of its behaviour with frequency. If the input is X and the output is Y, we could write Y(f) = H(f)X(f), where H is the transfer function – the relationship between the input and output signals. We write Y(f), H(f) and X(f) to show that the signals (which could be voltage or current) and the transfer function vary with frequency (f). The term “frequency response” often refers to graphs showing how the transfer function H(f) varies with frequency. Frequency response graphs often require a lot of data points, so it can be time-consuming and tedious to measure and calculate each point manually. Professional labs use test gear that can automate the measurement; many midto-high-end oscilloscopes now have this capability, but it is less common in older and lower-cost models. Alternatively, a separate signal generator and oscilloscope can be coordinated by software to automate measurements (eg, using the SCPI protocol). Not all readers will have even a basic oscilloscope, which could be used for manual measurements. It is possible to use a digital multimeter (DMM) to measure the gain response, but only if the meter can measure AC voltage over the required frequency range; basic models typically have limited capabilities in this respect. An audio millivoltmeter is a more appropriate instrument for this task. You would also need a signal generator to create the sinewave input. Many digital oscilloscopes these days come with built-in arbitrary waveform generators. If you do not have suitable test gear, there is a good alternative – use the “sound card” of a PC or laptop to generate the input signal and analyse the response. Fortunately, free software is available that provides this facility. These programs are mainly aimed at acoustic measurements (room acoustics, loudspeakers etc), not directly testing electronic circuits, but they can be used for this purpose. We will look at how to do this. By the way, we refer to the audio input/ output circuitry as a “sound card” because it used to be the case that computers had no such circuitry, so you needed an add-on card (circuit board) to provide these functions. These days, almost all computers have analog audio input/ output circuitry integrated onto their main boards, but the name has stuck. Before getting into the specifics of this process, we shall look at the basics of frequency response measurement, both the theory and some general practical considerations. Amplitude L ast month, we reached the end of our series on digital signal Time Time Fig.1: possible amplitude and phase shifts of some sample sinewaves. Practical Electronics | October | 2025 Fig.3: the LTspice circuit used to obtain the Bode plot in Fig.2. Fig.2: an example Bode plot produced by LTspice for the Fig.3 circuit. The upper plot is gain magnitude (frequency response), with the corresponding phase shift below in red. measurements) by applying a range of sinusoidal signals at the input and observing the output. If we feed a sinusoidal signal to a linear circuit, the output will also be a sinusoid at the same frequency – this must be so because a linear input-output relationship will not change the shape of the waveform. For something other than a sinewave, a different linear input-­output relationship may apply at different frequencies. This means that the component sine­ waves that build the waveform may not have the same amplitude or phase relationships relative to one another at the output. The waveshape will change if this happens, but the frequencies present will still be the same as in the input. This does not mean that it is impossible to measure frequency responses with inputs other than sinusoidal signals, but it makes things more complex, as we have to find the responses at individual frequencies from the combined response. This can be done using analysis based on the fast Fourier transform (FFT), which takes a waveform in time and finds its frequency content (or another type of Fourier transform). For a linear circuit with a sinusoidal input, the output waveform may be shifted in time, and may have a different amplitude (see Fig.1). When considering frequency responses, we usually represent the time shift as a phase shift, ie, the shift as a proportion of the cycle time (1/f) of the waveform at the frequency of interest (f) expressed in degrees (the full cycle time is 360°). For example, at 1kHz, a shift of 0.25ms relative to the input is a quarter of the cycle time of 1ms, so it’s a phase shift of 90° (360° ÷ 4). The basic manual approach to measuring a frequency response is to display the input and output waveforms on a dual-channel oscilloscope, and measure the amplitudes and phase shift, as shown in Fig.1. The gain of the circuit is calculated by dividing the output amplitude by the input amplitude. The phase shift is obtained by measuring the time difference Fig.4: the response of an RC circuit when a sinusoidal input signal switches on abruptly. Practical Electronics | October | 2025 between equivalent points on the two waveforms and converting this to the phase value. Whatever method is used to produce the data, the most common way of presenting frequency response data (simulated or measured) for electronic circuits is the Bode plot: graphs of gain in decibels and phase shift in degrees against a common logarithmic frequency axis (see Fig.2). The gain in decibels is 20log10(Vout ÷ Vin) for voltage signals. For completeness, the Bode plot in Fig.2 is for the RLC circuit in Fig.3, but the point of Fig.2 is to show the typical presentation of a Bode plot, not to discuss details of the circuit. The Bode plot is named after Hendrik Wade Bode (1905-1982). His work on this, dating to the 1930s, related to obtaining straight-line asymptotic approximations of the gain and phase responses of systems to assist in calculating their stability. Today, the term is widely used to apply to fully detailed frequency response curves, not just the straight-line approximations. Phase and steady state A phase shift may be positive or negative, with a negative phase shift implying the output waveform changes after the input (sometimes called phase lag). A positive shift is the opposite – the output changes occur before the input (phase lead). A positive phase shift may seem impossible – it appears to indicate the circuit is predicting the future input to produce the output. If this actually happened, we would have a non-causal circuit, which cannot be physically implemented for real-time operation. However, a simple high-pass filter using a single capacitor and resistor has a positive phase shift. The solution to this potential dilemma is that frequency response is defined for a steady state, that is, under conditions where the input has been applied for a long time (ideally for all time) and the circuit has settled into a stable input and output waveform. Under these conditions, the output can have any positive offset in phase without the system needing to be clairvoyant. If we apply a sudden input change (a step, or just switching on a sinusoid), the system will not be in a steady state 53 directly after the change. For example, Fig.4 shows a sinusoid applied to an RC high-pass filter just after switch-on. During the first half-cycle, the output (red trace) does not lead the input, but the circuit settles to a state where the output has a positive phase shift, similar to the plot on the right of Fig.1. A steady-state sinusoid is a requirement for there to be a single frequency at the input. In theory, steady state requires infinite time to achieve, but in practice, most circuits will reach a point that is close enough in a relatively short time. However, the time required for it to stabilise may still need to be taken into account when making measurements. The theoretical need for a steady state for a perfect sinusoidal input and output does not mean that we have to try to achieve this to make a frequency response measurement. It only applies to the case where we directly measure the amplitudes and phase shifts at individual frequencies (as in Fig.1). As indicated above, with appropriate mathematical processing of the signals, other input waveforms can be used. We will discuss this in more detail later. Complex numbers A key thing that Fig.1 shows is that there are two changes occurring from input to output – the phase shift and the amplitude change. If we write Vout = GVin for an amplifier at the frequency of interest, we can say that if G = 10 and the input is 1V, the output will be 10V. However, this does not tell us anything about the phase shift that is also part of the transfer function. This means that in Y(f) = H(f)X(f), we cannot simply say H(f) = G. The terms in Y(f) = H(f)X(f) have to encompass both the amplitude and the phase. Ordinary numbers have only one value, so we need something more. We can represent the phase and amplitude of a sinusoidal signal with amplitude A and phase Ф directly by writing it as A∠Ф, which is called the polar form. This is depicted in Fig.5 as a line of length A at an angle of Ф to the axis. The signal’s value is represented by the point at the tip of the arrow, which can also be expressed in terms of its Cartesian coordinates (x and y on a graph), also as shown in Fig.5. From basic trigonometry, x = Acos(Ф) and y = Asin(Ф). We could just use x and y as two ordinary numbers, but that is not an effective approach for mathematical manipulation. The way forward is to use complex numbers, such that the value of the signal in Fig.5 is x + jy, where j is √-1 (for some reason, mathematicians like to use i for this but engineers use j because they use the variable i to represent current). 54 Imaginary y jy Non-linear function y = f(x) j = √(–1) A Linear approximation Operating point Φ x x Real Fig.5: the relationship between phasors and complex numbers. Fig.6: the linearisation of a transfer function at an operating point. Ch1 Out Signal generator In Device under test Out Oscilloscope (or DMM) Ch2 Fig.7: typical lab test equipment for measuring frequency responses. This is called the rectangular form of the number. In x + jy, the x is referred to as the real part and jy as the imaginary part (the square root of minus one does not exist as an ordinary number). Complex number representations of sinusoidal signals are referred to as phasors. Using complex numbers, two values can be manipulated together in calculations and algebraic manipulations without them interfering with each other (mathematicians call this orthogonality). The use of complex numbers is fundamental to the theory of frequency-dependent impedance and the transfer functions of components and circuits. You often see transfer functions denoted as H(jω) and H(s), where ω is the angular frequency in radians (ω = 2πf ), and s is the complex frequency variable used in Laplace-transform-based circuit analysis. The fact that the transfer function is based on complex numbers means that when we refer to just the amplitude part of the gain, we should call it “gain magnitude” or just “magnitude”. Bode plots are often labelled this way, but not always. Using the Pythagorean theorem, the magnitude of the signal in Fig.5 is A = √x2 + y2. Also, Ф = tan-1(y ÷ x). Circuit analysis based on complex numbers uses advanced mathematics that not everyone will want to study. However, even if it is not for you, it is worth being aware of the basic ideas and notation as you will come across it in IC datasheets, other technical documents, software menus or help documentation. For example, LTspice can display AC analysis results in terms of real and imaginary parts. Fortunately, you do not need a deep knowledge of advanced mathematics to measure or simulate frequency responses and interpret the results presented as Bode plots. Often, the gain plot will be the main interest – for example, to decide if the gain of an amplifier is flat enough over the frequency range of interest, or if the cut-off performance of a filter is good enough for the application. However, phase is important too, as it influences signal integrity, distortion and system stability. An aside on LTspice An aside for those interested in the finer details of LTspice: you may have noticed the fact that the gain magnitude and phase plots in Fig.2 are in separate plot planes. The default for an AC analysis is to plot both on the same plane, with the phase axis on the right and the phase graph having a dotted line in the same colour as the gain. When you add the data points, as in Fig.2, the phase line can look quite messy. I do not know of a way of changing the default phase plot to use a solid line, a different colour, or the left-side axis. The traditional form of the Bode plot is to use separate graphs for gain magnitude and phase, but for software tools such as LTspice, it makes better use of screen space to display both together. Plotting the gain in separate panes in LTspice is simple if you just plot the output twice and switch off one Practical Electronics | October | 2025 Using a suitable DMM or audio millivoltmeter will require you to switch the connection between input and output (unless you have two), and will only give you gain magnitude values. For automated measurements, the signal generator may be built into the oscilloscope, or separate instruments could be controlled by a PC or linked by some proprietary system. For radio-frequency work, a vector network analyser (VNA) may be used rather than an oscilloscope, and may also measure other parameters, such as input impedance and reflection coefficient. In this article, we are focusing on lower (primarily audio) frequencies. Signals for measurement Fig.8: examples of sinusoidal signal distortion. The axis units are arbitrary. of the plots in each plane off, by right-­ clicking on the relevant axis. However, the phase still has a dotted line, and the axis on the right. The solution is to plot ph(V(out))*1° instead (assuming the output signal is named out). This uses LTspice’s “waveform arithmetic”. To do this, you need to click in the pane and “Add Traces”, then type in the expression. Just plotting ph(V(out)) will not show negative phase values correctly, because the trace plotter is interpreting the data as a magnitude, which is what would usually be done for gain in a Bode plot as the default for an AC analysis. This can be fixed by right-clicking the axis and selecting Cartesian rather than the default Bode representation (polar form). This plots the real and imaginary parts of the data (rectangular form). The LTspice help on waveform arithmetic tells us that the ph(x) function will “return a complex number with the real part equal to the phase angle of the argument and the imaginary part equal to zero”. We can right-click and switch off the imaginary axis (on the right) – as just stated, all the data points are zero. The numbers on the left axis are written with “r” (for real number), which is not ideal. Including *1° in the waveform expression fixes this. Simulation and measurement The theory of circuit frequency response is based on an assumption of linearity, but real circuits are not perfectly linear, so we need to consider the implications of this. When mathematically analysing circuits such as transistor amplifiers, it is common to use what are called smallsignal models to approximate non-linear Practical Electronics | October | 2025 relationships (eg, y = f(x) in Fig.6) to linear ones at an operating point (specific value of x). As long as the value of x does not deviate much from the operating point (hence, “small” signal) the linear approximation is good. LTspice (and other SPICE simulators) can simulate the frequency response of circuits using AC analysis (as in Fig.2). This works by calculating the operating point and linearising the current-voltage relationships of nonlinear devices such as transistors. The linear models allow very fast calculation of data points on the frequency response without having to simulate using input waveforms. AC analysis in SPICE does not work in all situations, such as sampling and switched circuits. LTspice is also able to perform a frequency response analysis (FRA) simulation, which follows a process much closer to what might be used for practical measurements. It performs transient simulations with sinusoidal waveforms over a range of frequencies and calculates the frequency response from the resulting signals. This is more complicated to set up than AC analysis. We used this approach for a sampling circuit in the DSP series (December 2024) and covered the basics of LTspice FRA in the March and April 2024 issues. The basic lab setup for frequency response measurements is shown in Fig.7. This can be used for manual or automated measurements. For manual measurements, set the frequency on the signal generator and measure the signal amplitudes and time difference using the oscilloscope. Calculate the gain and phase shift as detailed above with reference to Fig.1. Move on to the next frequency and repeat. Analysis and simulation based on small signals is very useful, but for measuring real circuits, we cannot just apply an arbitrarily small signal because we want the measurement to apply to realistic usage. Furthermore, if the signal levels are too low, noise in the circuit and/or test instruments will reduce the quality of measurements. For the circuits we are typically interested in for frequency response measurement, such as amplifiers and filters, good circuit design should provide decent linearity over normal operating signal levels. Typically, we want to apply the largest signal that is within the circuit’s safe operating range, but which avoids any significant distortion of the output signal. Common forms of distortion are clipping, where the maximum signal amplitude is exceeded, and slew-rate limiting, where the circuit cannot change its signal level fast enough to reproduce the waveform shape correctly. These are illustrated in Fig.8. Clipping results in sinewaves with flattened peaks, while slew rate limiting makes the waveform more triangular. Relatively severe cases of distortion can be spotted on an oscilloscope display, but may not be easy to see at lower levels. Some measurement systems will be able to measure distortion directly as well as plot frequency responses. Many oscilloscopes can display signal spectra, which can be used to help detect distortion. A sinusoidal output should have one peak at the input signal frequency. Other significant peaks, particularly at multiples of the input frequency, indicate distortion. As a sinewave is clipped more and more, it approaches a square wave, which will have high odd (3f, 5f etc) distortion products. The maximum signal level at different frequencies may be different, and can be determined by different characteristics of the device under test. For 55 manual measurements, we could change the signal level for each measurement, but automated measurements may use a fixed level for all frequencies. Clipping can occur due to either the input or output level being too high. Exceeding the maximum input level is likely to be the limiting factor where circuits attenuate the signal (large input for small output); for example, filters in the stop-band frequencies. Exceeding the maximum output level is more likely for circuits that amplify in all or part of the frequency range (amplifiers, filters in the passband). The maximum rate of change of a sinusoidal waveform occurs as it passes through zero. Increasing either amplitude or frequency increases the maximum rate of change, so slew-rate limiting is more likely for relatively large, high-frequency signals. This could occur, for example, for amplifiers and high-pass filters at high frequencies, or low-pass filters close to or just past cutoff. Clipping problems can also occur in the measurement instrument input, which will invalidate the measurement. Ideally, you will get a warning about over-range/overload or clipping if this occurs. The above discussion mainly applies to active circuits (eg, using transistors and op amps), but you may want to measure the frequency response of passive circuits (containing a combination of resistors, capacitors and inductors). In general, typical test set-ups are unlikely to exceed component ratings, but check if this may be a possibility. Also be aware that resonant passive circuits (those containing LC-based filters) may produce voltages greater than the input voltage. Several types of input signal could be used for frequency response measurement – examples are shown in Fig.9. The most straightforward is a stepped sinewave. As already discussed, this is what is used for manual measurements, but of course, it can be automated too. Higher-resolution measurements (more frequencies and more accurate response data) will take longer. Time is required for the output to stabilise (approximate steady state) each time the frequency changes. Alternatives to stepped sines include swept sinusoids (also called chirps), where the frequency varies continuously, or wideband random noise (eg, white noise). These input waveforms contain frequency content at all frequencies of interest, so the output frequency content can be analysed to obtain the frequency response. This is done by processing the 56 Fig.9: signals that can be used for measuring a frequency response. Fig.10: the same frequency response plotted with 10 points per decade (upper) and 500 points per decade (lower). waveforms for the whole sweep, or an appropriate duration of noise, using calculations based on the fast Fourier transform (FFT). Logarithmically swept sine measurements tend to be much faster than stepped sine, but the measurements may be worse, particularly at higher frequencies due a lower signalto-noise ratio (SNR). In theory, an impulse can be used as the input waveform to obtain the frequency response. Again, an FFT is used to obtain the frequency response. An ideal (analog) impulse is not feasible, but a short high-amplitude pulse will approximate it. There are practical difficulties – the pulse has to be very short, and therefore causes a small output signal, and hence poor SNR unless it has a very high amplitude, which may not be feasible, or could cause non-linear behaviour. For digital filters, impulses may be usable if a single non-zero input sample can be reliably created. We discussed impulses in the DSP Practical Electronics | October | 2025 Fig.11: a linear plot of the same response shown at the bottom of Fig.10. Fig.12: LTspice phase plots of the same network. The upper plot wraps the phase at ±180°, while the lower doesn’t. series, particularly in the August 2024 and April 2025 issues. Data points and plots Bode plots usually use a logarithmic frequency axis. This means that frequencies a decade apart (factor of ten) are equally spaced on the axis. For example, there is the same distance from 100Hz to 1kHz on the axis as there is from 1kHz to 10kHz (as in Fig.2). The number of data points to use is often specified in terms of points per decade, or sometimes points per octave (an octave is a doubling of frequency). Fig.2 has 10 points per decade, shown by the dots on the graph. LTspice’s plotting software has interpolated the points between these to draw a smooth curve. When making manual measurements, how do you choose which frequencies to measure at if you want equally spaced values on a logarithmic plot? Each measurement frequency (f) is a fixed scaling factor (s) larger than the previous one, ie, fn+1= sfn, when n is the measurement number in order of increasing frequency. Practical Electronics | October | 2025 For p points per decade, if we multiply s by itself p times, we get 10; that is, p sp= 10, so s = √10. For example, if we wanted four points 4 per decade, we have s = √10 = 1.7783. Our frequencies in each decade are the start frequency multiplied by 1, 1.7783, 1.77832 and 1.77833. For example, starting at 100Hz, we would have 100Hz, 178Hz, 316Hz and 562Hz. Multiplying the final value by 1.7783 gives us 1000Hz – the start of the next decade. We can calculate all the frequencies to measure by starting at the lowest frequency we need and multiplying by s until we get to the maximum frequency to use. Similarly, for p points per octave, the p scaling is s = √2. The number of octaves per decade (N) is 2N = 10 (N doublings is equal to ten). So, N = log(10) ÷ log(2) = 1 ÷ log(2) ≈ 3.322. Insufficient data points can lead to misleading plots. Fig.10 shows the gain response for the circuit in Fig.3 with R1 changed to 0.1Ω. This low resistance results in a very sharp peak, which requires a lot of data points to plot accurately. The upper plot uses 10 points per decade, which is insufficient. The lower plot uses 500 points per decade. For manual measurements, it can be useful to do an initial run with relatively few data points and then only make closely spaced measurements where the response changes rapidly. For steppedsine automated measurements, using a lot of data points may take a long time, but at least you can do something else while you wait. For measurements based on FFTs, the number of data points will depend on the number of waveform samples (in time) and the FFT algorithm. Usually, the number of data points in the frequency response plot will match the number of waveform samples, and will be evenly rather than logarithmically spaced. Typically, the number of points will be a power of two and be a large number (eg, 16384, 32768 or 65536), so likely to be sufficient for accurate plotting. Fig.11 shows the same data as the lower plot in Fig.10 plotted using linear rather than decibel gain. We are not really able to see the fact that the circuit is a low-pass filter because the peak dominates the plot. This is a relatively extreme case, but illustrates the usefulness of decibel plots or logarithmic axes in showing details at different scales. Linear axes are more appropriate for plotting smaller ranges. The symmetry of the sinewave means that just by looking at two waveforms on an oscilloscope, it is impossible to distinguish between phase shifts of +180° and -180° (they both invert the waveform). Similarly, shifts above 180° in one direction look the same as smaller shifts in the opposite direction (eg, +270° and -90°). Simulations and more advanced measurements may be able to correctly obtain phase values beyond ±180°, but it is still common to plot values in the ±180° range (called phase wrapping). Simulation and measurement may give you the option to plot phase wrapped to ±180° or unwrapped to the full value; for example, see the LTspice plots in Fig.12, which is for an RLC circuit formed from two copies of the filter in Fig.3. Circuits with relatively large delays may have very large phase shifts at high frequencies, and wrapped plots can show a large number of jumps between ±180°. Sound card based measurements There are many PC programs, including some free ones, designed for audio applications such as room acoustics measurement, characterising loudspeakers and other audio devices/systems. These use the PC’s sound inputs and outputs to make measurements. 57 One of their main capabilities is to make measurements to help optimise the acoustics of a room or studio and find the best locations for speakers and listening. This requires a calibrated microphone or sound pressure level (SPL) meter, which may be analog or USBconnected in a setup like in Fig.13. These programs may provide standalone signal generator and oscilloscope tools, and of course, they can measure and plot frequency responses, as this is a key task in room audio characterisation. They may also be able to measure other parameters, such as signal distortion and component impedance. Fig.14 shows a simple setup that could be used to test an electronic circuit. This is similar to the arrangement in Fig.7, where the sound card’s right (stereo) channel is used as the signal generator output, and the two input channels are used to measure the input and output signals. Other arrangements may be used; for example, the left I/O channels may be connected together (loopback, as in Fig.13) to facilitate timing measurements. Direct measurement of the input signal may not be needed if the sound card is calibrated first using a direct connection (R out to R in) without the device under test (DUT) present. Compared with using lab test gear, sound-card-based measurements have limitations. The input and output voltage ranges are limited and may vary significantly, depending on the sound card, particularly between built-in standard sound cards and the external ‘professional audio’ USB devices often used by musicians. The frequency range is limited to the audio spectrum (20Hz to 20kHz), although some sound cards with higher (“HD” or “studio”) sampling rates (eg, 192kHz) may facilitate higher frequency measurements. Sound cards are designed for audio signals only, and are usually AC coupled, so DC measurements are not usually possible. Sound cards are not designed as measurement instruments, so are not calibrated. The software usually provides calibration procedures to overcome this. One potential advantage is that they may have greater signal level resolution (eg, 16 or 24 bits) than is typical in oscilloscopes (8 to 12 bits). The much lower sampling frequencies used by sound cards, compared with what is expected of an oscilloscope, makes this possible. Measurements may require two inputs (using stereo, as shown in Figs.13 & 14), which should not be a problem on many PCs, but laptops commonly only have a single channel microphone input, 58 Audio system PC Sound card out Main R speaker output L R Aux/ L Line in Line R L Optional loopback connection Line L in R Microphone/ SPL meter Speakers Alternatively, connect the SPL meter via USB USB Fig.13: a typical setup for acoustic frequency response measurement using a sound card. The audio system can be directly measured by attenuating the speaker output signals to levels that are suitable for feeding into the sound card line-in terminals. Only try that if you really know what you are doing! Fig.14: a simple setup for testing an electronic device using a sound card. PC Sound card Line R out L In Device under test Line L in R PC Sound card Out In Line R out L Out Loopback connection In Line L in Device under test R Out USB In USB Fig.15: signal conditioning to match the sound card with the DUT signals. Input signal conditioning Output signal Out conditioning Possible USB connection – eg, for MCU dev board often combined with a stereo output via a TRRS (tip-ring-ring-sleeve) headset socket. An external (eg, USB) audio interface can also be used. The arrangement in Fig.14 may work in some cases, but often, the signal levels and drive capabilities for the DUT and sound card I/O may not be compatible. This will require signal conditioning circuits on either or both the input and output of the device (see Fig.15). Coming up We will continue to discuss measuring frequency responses with a computer sound card in the second article in this PE series, published next month. Practical Electronics | October | 2025