Ceramic filters are electronic components used as the bandpass element for Intermediate Frequency (IF) amplifier stages of an FM radio. Ceramic filters can also be employed for other purposes such as AM and TV IF stages, but this page concentrates on their use in FM tuners and receivers. The general discussion here applies to these other applications as well.

Ceramic filters are available from many manufacturers, but usually share several characteristics in common. They are three lead rectangular devices that resemble ceramic capacitors except for the three leads. The leads are on 1/10 inch spacing. They can be various colors (although brown is the most common) and heights, although the middle picture below depicts a Toko filter which is usually red and lower profile.

The three pins from left to right are input, ground, and output. The devices are polarized so make sure that you get the input pin connect.

The electrical function that ceramic filters perform is that of a passive bandpass filter centered on 10.7 MHz. A bandpass filter passes its center frequency, while rejecting frequencies to each side of the center. The quality, or "Q" factor of a bandpass filter is a measure of how sharply it rejects the frequencies to each side of the center frequency. The Q of FM ceramic filters is not specified directly, instead it is stated as the bandwidth at which the filter has rolled off by 3 dB above and below the center frequency. For a "150 kHz" ceramic filter, then, the signal would be rolled off 3 dB at 10.625 MHz and 10.775 MHz. This is shown in the picture below:

This figure needs a number of comments:

• The action of a tuner "mixer" stage translates all FM stations down to a frequency of 10.7 MHz. Therefore, stations on adjacent frequencies would appear at 10.5 and 10.9 MHz, stations on alternate frequencies (two frequencies away) would appear at 10.3 and 11.1 MHz, etc.
• For those wanting to know the "Q" of this particular 150 kHz ceramic filter, the formula is center frequency divided by the bandwidth, or 10.7 MHz / 150 kHz = 71.333. Q is a unitless quantity.
• The center of the bandpass response on the graph has a normalized amplitude of 0 dB (no loss). Ceramic filters are passive devices, and have some associated insertion loss. This insertion loss is usually in the range of 2 to 10 dB - and must be taken into account by the IF amplifier and feedback system of the tuner.
• The vertical scale of this figure is greatly exaggerated - showing single dB's of rejection. This might lead you to believe that the selectivity from a single filter would be fantastic, but the exaggerated vertical scale enhances the appearance of selectivity.
• The group delay of this particular filter is displayed on the same graph. A flat group delay is of special interest to audiophiles, because it will lower the total harmonic distortion of the tuner. DXers will be less interested in flat group delay.

The real story on tuner selectivity is better shown by a graph that is wider in both frequency and amplitude:

• The specification usually used for tuner selectivity is "alternate channel selectivity". The formal definition of selectivity involves the background interference received on channel when a signal on the specified channel is modulated 100% by 400 Hz. For the purposes of this discussion - I will just look at the response as I would that of a bandpass filter, because I am more comfortable working in that terminology than I am talking in terms of true selectivity. The ceramic filter is a passive bandpass element, which passes frequencies in a narrow band called the "pass band" centered at about 10.7 MHz, and rejects all others in two regions called the "stop bands", above and below the passband in frequency. But - as a bandpass element, it is not perfect. The response is not perfectly centered on 10.7 MHz. In most cases, it does not have infinitely steep slopes at the edge of the pass band, and the ultimate rejection in the stop bands is not infinite. Looking at the figure, it shows the traces of the bandpass characteristics of four commonly used FM ceramic filters - 280, 230, 180, and 150 kHz. The 280 kHz trace is shown in blue. If the tuner utilized only one of these 280 kHz filters, the alternate channel below the frequency being received would only be rejected by 38 dB. Above the pass band, the rejection is 52 dB which is somewhat better, but the customary way of specifying numbers is to rate the worst case. Looking at the red trace, which corresponds to the narrow filter, the high side rejection is only 48 dB, even though the low side rejection is much better. This asymmetrical filter response is typical, in fact early Toko filters were known for having poorer high side rejection than low side - a problem they have subsequently fixed. These stop band rejection numbers are lackluster - any tuner that is better than a portable radio will probably employ more than one filter to improve the stop band rejection. Portable radios will almost always have additional transformer based IF stages enhancing the stop band rejection.
• The situation changes dramatically when you talk about adjacent channel rejection instead of alternate. Adjacent channels are only 200 kHz away from the center frequency of the station tuned. Looking at the blue filter trace, which corresponds to a 280 kHz filter, the adjacent channel rejection is only 10 dB, which is almost nothing at all. In fact - stations on adjacent frequencies are almost certainly much weaker than local stations, so obtaining satisfactory reception on them requires a much higher degree of rejection. The 150 kHz narrow ceramic filter response shows that the adjacent channel rejection is much better than with the 280 kHz filter: 30 dB instead of 10 dB. This is probably insufficient, however, for the critical requirement of separating out a weak station next to a local. The real improvement will come when the narrow filters are cascaded. Only a small change in slope of the bandpass characteristic will produce a large change in rejection. On the other hand - real filters have some variation in their center frequency. It makes no sense to have highly selective filters if the center frequency is off, and the response characteristic therefore includes large chunks of the adjacent station's signal. For high selectivity, it is crucial to match multiple ceramic filters in center frequency!
• Not all "narrow" ceramic filters are equally good! The best measure of their ability to reject adjacent stations is their 20 dB bandwidth, not 3 dB bandwidth! This specification tells you how fast the filter rolls off, instead of just the bandwidth on the center frequency. It may well be that a 180 or even 230 kHz filter from one manufacturer actually rejects adjacent channels better than a 150 kHz filter from another. See the specification table below for the best filters! Also - be aware that the bandwidth specification can have a tolerance of close to 20%. It is possible to get a "bad" batch of filters with unusually wide bandwidth. Cherish a batch of good filters when you get them!
• Another thing that you should notice about the figure above: the ultimate stop band rejection of ceramic filters bottoms out at about 40 dB. Therefore, stations from even at the other end of the dial will only be 40 dB down with a single ceramic filter! This may be adequate in a $5 clock radio, but is seldom adequate for serious listening. The result of using a single ceramic filter with no other IF bandpass elements will be a cacophony of local stations mixed together, when the receiver is near several strong stations. Fortunately, the stop band rejection improves with each additional filter. The ultimate limit of stop-band rejection will be parasitic bleed through on the PC board - something that is best combatted by using at least a two sided board with a good ground plane.
• Wider bandwidth ceramic filters are needed if you want to receive any of the subsidiary services on FM sidebands, such as SCA music, RDS, or reading for the deaf for example. The narrow ceramic filter modification may disable these features. Wider bandwidth is also required for the hybrid In Band On Channel (IBOC) digital mode of iBiquity's system.

If you modify tuners, you will need some method of testing ceramic filters. The figure below is taken from a Murata data sheet (I have corrected a mistake where they swapped the output and ground pin numbers):

While this method will work, I am fortunate to have access to an Agilent 8753E, which allows me to get a good visualization of the filter characteristics. A little modification to the diagram above is needed, however. Instead of the AC voltmeter, I connected the high impedance input of a non-inverting op amp stage to the filter output:

This allowed me to match the 50 ohm input impedance of the network analyzer to the 300 Ohm impedance of the filter. A THS4304 has a gain / bandwidth product of 3 GHz, so its effect on the characteristics of the filter is negligible. The gain of the op amp stage is 2 at the amplifier output, but the effect of matching / terminating resistors results in an overall gain of 1 at the network analyzer input. Therefore, this circuit can be used to measure the insertion loss of the filter.

A network analyzer plot of a good filter is shown on the red trace below:

Occasionally, a filter showed a response that was not symmetrical, as shown in the green trace. Perhaps I am obsessing a little bit about the shape of the bandpass characteristic. Ideally, it would be perfectly flat across the top, then have a perfectly vertical dropoff to zero on each side of the pass band. As noted above, this is not the case. In the real world, you at least would want the passband to do the same thing on either side of the center frequency. This way, the frequency deviation would be symmetrical on each side as the station deviates +/-75 kHz. The nature of frequency modulation makes it immune to amplitude variations in ideal realm, but in the real world amplitude variations will affect the audio to a degree. So I flag these devices as undesirable, and do not use them.

The next aspect of filter response to consider is center frequency. For those of you who have digital tuning, you are stuck with having to find filters that are as close to 10.7 MHz as possible. This is because the tuning voltage will set up the varactor diodes in the front end of the receiver for an exact 10.7 IF. If you are really ambitious, you might see if you can intercept the tuning voltage before it goes to the front end and add a trimming circuit. This will have two benefits:

• You will be able to use just about any frequency of filter, whether they are 30 kHz or more off of center.
• You will have a fine-tuning adjustment that will be useful in closely packed station situations.

The red trace above shows a nice, symmetrical response that is centered slightly higher than 10.7 MHz in frequency. As long as the other filters in the IF stage are matched to this center frequency, you will be fine - especially if you have an analog tuner. The slightly high IF frequency will result in a miniscule difference in dial calibration - and that is all.

In order to really understand what is going on with the filter near its center frequency, it is necessary to increase the resolution near the center frequency:

This figure shows a key fallacy of the AC voltmeter method of test selecting filters. This filter is not very symmetrical. The vertical scale of the diagram is 0.5 dB, and the passband is skewed to the right. The AC voltmeter method of finding the peak will yield a center frequency of about 10.745 MHz. If this filter is used to match other 10.745 MHz ceramic filters in an IF stage, a portion of its skewed response will amplify unwanted adjacent stations. Marker 2 shows that the -3 dB point on the low end of the passband is 100 kHz from the amplitude peak, while marker 3 shows the upper end of the passband is only 66 kHz from the amplitude peak.

Rather than characterize this filter based on the peak from the AC voltmeter, I think it should be characterized by its -3 dB points as shown here:

Here, all references have been changed from the peak (marker 1) to the center frequency (marker 4), which is 17 kHz lower. Doing this places the filter within spec for the center frequency (about 10.727 MHz). The -3 dB points are +/-84 kHz, which means this filter is a 168 kHz wide filter. That is within spec for a 150 kHz +/- 30 kHz bandwidth filter. Again - I would mark this filter as unacceptable for use in high end tuners, but it does illustrate several aspects of filter performance.

The following manufacturers make ceramic filters. Of these - two are commonly available to hobbyists through Digi-Key - Murata and Toko

• Hong Kong Crystal
• Kyocera ceramic resonator only, but made ceramic filters at one time
• Languang Electronic Company
• Murata
• Nanjing Sunny
• Shoulder
• Technical Crystal
• Token
• Toko
• Zhejiang Jiakang

Manufacturers can sometimes be identified by color:

• Murata filters are brown, and nearly square - or slightly wider than tall.
• Toko filters are red, and somewhat shorter than filters usually are.
• Kyocera ceramic filters (now discontinued) are both brown and short.
• Shoulder ceramic filters are tan, and have shiny cases.
• Technical Crystal ceramic filters are dark brown.

Murata is the most common manufacturer, and is very easy to get from Digi-Key. There are two tables associated with Murata - a legacy part number table and a new part number table. The legacy part number table is included because there are a tremendous number of older legacy filters still available through distribution and in tuners - manufactured before a part number standardization initiative in 2001. The tables are sorted by typical 20 dB bandwidth. The more desirable filters for broadband use are near the top, and the more desirable filters for DX use are near the bottom of each table.

NOTE: Many manufacturers follow the center frequency dot convention. Murata uses the location of the dot to differentiate part markings, as described in the table below.

Toko is another common manufacturer, and is very easy to get from Digi-Key. These filters are generally shorter than Murata, and are easily identified by their red color. Some DX'ers have reported that high-side selectivity is not as good as low side. A look at their data sheets reveals close in response that looks symmetrical, so they have either fixed their process or the high-side rejection problem only shows up on wider plots.

A number of Chinese manufacturers have cooperated in adopting a standardized part numbering system for FM ceramic filters. Several provide filters in standard and low profile sizes. Often times, the low profile products are also low loss / better selectivity. Sort of an attempt to put all of the "good" characteristics into a premium "A10" part number. The following table is compiled from a list of typical specifications from these manufacturers. It may be very difficult to identify the exact manufacturer of a ceramic filter by the part marking - as these are commodity items. The primary value of this table is to summarize device characteristics from the part marking. While SNR is not a Chinese manufacturer, it still adheres to the Chinese numbering scheme.

Some of these manufacturers no longer make ceramic filters, but I have included them in the list because they did at one time and their products are still found in many FM receivers.

Manufacturer legend for table above:

One manufacturer - Sunny Electric - deviates slightly from the standard, and so they get their own table:

• Ceramic filters and resonator
• Bob's Filter Corner on FMTunerInfo.com
• Improving FM Reception Better FM
• IF Filters by Mike Hawk
• Better Selectivity and Sensitivity From Virtually any FM Radio by Bruce Carter