Digital IF Overview - Spectrum Analysis

One of the most profound changes in spectrum analysis since the 1980s has been the use of digital technology to replace the analog circuits in previous instruments. With the introduction of high-performance analog-to-digital converters, the latest spectrum analyzers can digitize the input signal earlier in the signal path than was available only a few years ago. This change is most evident in the IF portion of the spectrum analyzer. Digital IF1 has greatly improved the performance of spectrum analyzers, greatly improving their measurement speed, accuracy, and ability to measure complex signals using high-performance DSP technology.

Digital Filter

Keysight ESA-E series spectrum analyzers use a portion of digital intermediate frequency circuits. Traditional analog LC and crystal filters can only achieve a resolution bandwidth (RBW) of 1 kHz and above, while digital technology can achieve the narrowest bandwidth of 1 Hz to 300 Hz.

Figure 3-1. Digital implementation of 1, 3, 10, 30, 100, and 300 Hz resolution bandwidth filters in the ESA-E Series spectrum analyzer.

As shown in Figure 3-1, the linear analog signal is down-converted to an intermediate frequency of 8.5 kHz and passed through a bandpass filter with a bandwidth of only 1 kHz. The intermediate frequency signal is then amplified, sampled and digitized at a rate of 11.3 kHz.

Once the signal has been digitized, it is subjected to a Fast Fourier Transform. In order to perform the conversion on the appropriate signal, the analyzer must be fixed tuned (not swept), i.e. the conversion must be performed on the time domain signal. Therefore, when we select a certain digital resolution bandwidth, the ESA-E series analyzer increments the local oscillator frequency in 900 Hz steps instead of continuously sweeping. This stepped tuning can be observed on the display, which updates in 900 Hz steps when the digital processing is completed.

We will see later that some other spectrum analyzers and signal analyzers (Keysight X-Series analyzers) use fully digital IF technology, that is, all resolution bandwidth filters in the instrument are implemented using digital technology.

A key benefit of the digital processing used in these analyzers is that they can achieve bandwidth selectivity of approximately 4:1. This selectivity is achieved even with the narrowest filters, and can be used to resolve signals with very close frequencies.

Let's do the same calculation for a digital filter. A good digital filter selectivity model is a Gaussian-like distribution:

Where H(∆f) is the filter edge roll-off in dB.

Δf is the frequency offset in Hz from the center frequency and α is the parameter that controls the selectivity. For an ideal Gaussian filter, the value of α is equal to 2. The swept RBW filter of Keysight's spectrum analyzers is based on a quasi-Gaussian model with α = 2.12, resulting in a selectivity ratio of 4.1:1.

Strictly speaking, once a signal is digitized it is no longer an intermediate frequency (IF). The signal at this point is represented by digitized values. However, we use the term "digital IF" to describe the digital processing technology that replaces the analog IF used in traditional spectrum analyzers.

Substituting the values ​​in the example into the formula, we get:

At a frequency offset of 4 kHz, the edge of the analog filter drops to -14.8 dB, compared to -24.1 dB for the 3 kHz bandwidth digital filter. Because the digital filter has this superior selectivity, it is better able to distinguish signals that are very close in frequency.

Full digital IF

Spectrum analyzers such as the Keysight X-Series combine multiple digital technologies for the first time to achieve a fully digital IF, which brings great benefits to users. The combined use of FFT analysis for narrow spans and swept frequency analysis for wide spans optimizes the scanning process, allowing measurements to be completed as quickly as possible. Structurally, improvements in analog-to-digital converters (ADCs) and other digital hardware have enabled the location of the ADC to be closer to the input of the spectrum analyzer.

Let's first look at the all-digital IF structure block diagram of the X-Series signal analyzer, as shown in Figure 3-2.

Figure 3-2. Block diagram of the all-digital IF architecture of the Keysight X-Series signal analyzer

In this structure, all 160 resolution bandwidth filters are implemented digitally, but there are analog circuits before the analog-to-digital converter: first, there are several stages of down-conversion, followed by a pair of single-pole pre-filters (one is a local oscillator filter and the other is a crystal filter). The pre-filter here is the same as the analog intermediate frequency, which is used to prevent the subsequent process from further amplifying the third-order distortion. In addition, it can also achieve dynamic range extension through automatic scaling. The output of this single-pole pre-filter will be connected to the autorange detector and anti-aliasing filter.

As with any FFT-based IF structure, an antialiasing filter must prevent aliasing (i.e., out-of-band aliased signals from being sampled by the ADC). This filter has multiple poles and therefore has a large group delay.

Even a fast rising RF pulse that is downconverted to an IF frequency will experience a delay of more than three ADC clock (30 MHz) cycles when passing through this antialiasing filter. This delay gives the analyzer time to identify an approaching large signal before it overloads the ADC. The logic controlling the auto-amplification detector reduces the gain of the signal before it reaches the ADC, thus preventing clipping. If the signal envelope is at a low value for a long time, the auto-amplification circuit will increase the gain accordingly, reducing the effective noise contribution at the input, and the digital gain after the ADC will also change accordingly to compensate for the change in analog gain before the ADC. The result is a "floating point" ADC with a very wide dynamic range when the auto-amplification function is enabled in swept mode.

Figure 3-3 depicts the sweeping behavior of the X-Series analyzer. The single-pole prefilter allows the gain to be very high when the analyzer is tuned far from the carrier frequency, and as it gets closer to the carrier frequency, the gain decreases and the ADC quantization noise increases. The level of this noise depends on the frequency of the signal from the carrier, so it looks like a staircase phase noise. However, phase noise is not the same as this AAC noise. Spectrum analyzers cannot avoid phase noise, and reducing the prefilter bandwidth can reduce the AAC noise at most carrier frequency offsets. Since the prefilter bandwidth is approximately equal to 2.5 times the RBW, reducing the RBW will also reduce the AAC noise.

Figure 3-3. Automatic amplitude scaling keeps the ADC noise close to the carrier and below the LO noise or RBW filter response.

DSP IC

Let's go back to the block diagram of the digital IF (Figure 3-2). After the ADC gain is determined by the analog gain and corrected by the digital gain, a dedicated integrated circuit begins to process the signal samples. First, it separates the 30 MHz IF signal samples into I and Q paths with half the rate (15 Mpairs/s), and uses a single-stage digital filter with a gain and phase opposite to the single-pole analog pre-filter to give the I and Q paths a high-frequency boost. Then the I and Q signals are low-pass filtered by a linear phase filter that is close to an ideal Gaussian response. Gaussian filters are often used in swept spectrum analysis because they best compromise between frequency domain performance (shape factor) and time domain performance (response to fast scanning). As the signal bandwidth decreases, the I and Q signals may be extracted and sent to the processor for FFT processing or demodulation. Although FFT operations can cover frequency bands up to the 10 MHz bandwidth of the anti-aliasing filter, even with a narrow FFT span (such as 1 kHz) and a narrow RBW (such as 1 Hz), 20 million data points are required to perform FFT operations. Using decimation techniques for narrow spans can greatly reduce the number of data points required for FFT operations and increase calculation speed.

For swept frequency analysis, the filtered I and Q signals are converted into amplitude/phase pairs. In traditional swept frequency analysis, the amplitude signal is filtered through a video bandwidth (VBW) filter and sampled through a display detector circuit. The selection of logarithmic/linear display and decibel value per scale is done in the processor, so the signal can be displayed on the screen at any scale without repeated measurements.

Other video processing functions

The VBW filter is commonly used to smooth the logarithmic magnitude of a signal, but it has many other functions. It can convert the logarithmic magnitude to a voltage envelope before filtering, and convert it back the same way before display detection.

The ideal way to view the envelope shape of a pulsed RF signal in zero span is to display the filtered signal amplitude on a linear voltage scale. Logarithmic amplitude signals can also be converted to power signals (amplitude squared) before filtering and then back again. Power signal filtering allows the analyzer to give the same average response to signals with noise-like characteristics, such as digital communication system signals, as to CW signals with the same rms voltage value. An increasingly common application requirement is to measure the total power in a channel or over a frequency range.

In this measurement, the displayed data represents the average power during the time period that the local oscillator sweeps through that data point. The VBW filter can also be configured as an accumulator to average logarithms, voltages, or power.

Frequency Count

Swept spectrum analyzers usually have a frequency counter. It is responsible for recording the number of zero crossings of the IF signal and the frequency deviation relative to the known offset of the local oscillator during the remaining conversion process. If the counter can count once per second, a high resolution of 1 Hz can be achieved.