Comparing the PNA "Delay" Functions

The PNA has three Delay functions which are similar but are used in different ways.

1.  Group Delay format is used to display the Group Delay of a network. Group Delay is defined as:

-d(phi)/d(omega) -- where phi is radian angle, and omega is radian frequency.  

Since it is defined by a derivative, the value must be determined from an analytic function. However, the PNA makes discrete measurements, so we approximate the group delay by taking the finite difference:

-(1/360)*delta(phi)/delta(f) -- where phi is degree angle and f is frequency in Hz. The 1/360 does the proper conversion of degrees to radians and Hz frequency to radian frequency.  

From this we can see that, if the phase response of a network varies with frequency, then the Group Delay must vary as well.  In fact, many filters are specified by the variation of their Group Delay.  

If we measure the phase response of a lossless cable, it should be a straight line. But, of course, nothing is perfect.  The phase response will have a small amount of noise. This is due to trace noise of the PNA, and the loss with real cables or transmission lines, which causes a small amount of non-linear phase change with frequency.  So, if we look at the Group Delay of a cable, we will see a small amount of variation.  Also, if the frequency spacing is small enough when you make the measurement, the delta(f) in the denominator becomes very small, so the delay can have wide swings with just a little noise.

To overcome this issue, we sometimes add smoothing to a phase trace, which widens the effective delta(f), called the aperture, and provides a less noisy Group Delay response.  The Group Delay of a device is only valid for a given frequency aperture. Learn more about Group Delay.

2.  Electrical Delay function.  On many filters, the passband response is specified for a maximum value of "Deviation from Linear Phase".  When looking at the passband of a multi-pole filter, one sees the phase changing very rapidly. This makes it difficult to determine the linearity of the phase response.  The Electrical Delay function subtracts out a "LINEAR PHASE" equivalent to the delay time value computed as above.   When you use this function, you dial in the Linear Delay such that a CONSTANT PHASE SLOPE is removed from the phase trace, until the phase trace is mostly flat.  The remaining variation is the deviation from linear phase.  

To make this task a little less tedious, the PNA has a marker function called Marker =>> Delay. This function computes the Group Delay value at the marker position, using a 20% smoothing aperture, then changes the Electrical Delay value to this value. Obviously, if the phase trace is not perfectly linear, moving the marker and recomputing the delay will result in different values.  The phase slope added by the electrical delay function applies only to the current measurement. That is, each measurement (S11, S22, S12, S21) can have its own value of electrical delay. Learn more about Deviation from Linear Phase.

3.  Port Extension is a function that is similar to calibration. It applies to all the traces in a given channel.  It compensates for the phase response change that occurs when the calibration reference plane is not the same as the measurement plane of the device.  

Let's look at an example of a DUT that is mounted on a PCB fixture with SMA connectors. We can easily calibrate at the SMA connectors. But if we add the fixture to measure the board-mounted device, the apparent phase of the DUT is changed by the phase of the PCB fixture.  We use port extensions to add a LINEAR PHASE  (constant delay) to the calibration routines to shift the phase reference plane to that of the DUT. This is ONLY valid if the fixture consists of a transmission line with linear phase response, and this limitation is usually met in practice.  The main reason that it is NOT met is that there is mismatch at the SMA-to-PCB interface. This mismatch was not removed with the error correction because it occurs AFTER the SMA connector. Ripple can be seen on the display as signals bounce back and forth between the mismatch and the DUT. If the DUT is well matched, the ripple effect is very small. However, when we use Automatic Port Extension (APE), and we leave the fixture open (the DUT removed), the reflection is large and we see larger ripples.  That is why APE uses a curve fitting process to remove the effects of the ripple.  For best effect, the wider the IF Bandwidth, the better we can "smooth-out" the ripples with curve fitting.  Still, we are fitting a LINEAR PHASE SLOPE to the phase response, and thus we use only a single Port Extension Delay value to represent the phase slope.

The method used by older VNAs to get this same functionality was to add a mechanical line stretcher to the reference channel, which removed a fixed delay amount from the port. Port extensions give 1x the delay for transmission at each port, and 2x the delay for reflection, so it differs somewhat from Electrical Delay above, in that the math function depends upon the measurement being made. The signal passes twice through the fixture for reflection (out and back), but only once for each port on transmission.  For S21, the phase slope added is the sum of the port 1 and port 2 Port Extension Delay values.

The "User Range" APE function is used in cases where a fixture has limited bandwidth, perhaps due to tuning elements or bias elements.  In this case, the model of constant delay for the fixture over the whole bandwidth is not valid, so a narrower "User Range" of frequencies can be selected to compute the delay. Since the aperture is smaller, there is more uncertainty in the delay computation for port extension.  Also, for those who had been using the Marker =>> Delay function to estimate the delay, we added the "Active Marker" selection to APE, which works exactly the same as Marker->Delay. Learn more about Automatic Port Extensions.