First published August 2010.
Go to comments (1) on this article.

Bandwidth and other technical matters in FM broadcasting
Factors affecting FM quality

This short technical article discusses some of the parameters that define the performance of FM broadcasts. The article ends with a short bulleted list of how to comply with MED-RSM licence requirements and a whimsical look at the ultimate end to end FM system.

August 2010

About the FM system

FM broadcasts started in the U.S in 1939, but didn't become properly commercial until the 1960's. In New Zealand we had to wait until the late 1980's to get FM, but 20 years later we have the highest density of FM stations per capita of anywhere in the world. With all the emphasis on things digital now, FM broadcasting has held out. The system is capable, with attention to detail, of providing very good quality and some have stated that FM audio quality is superior to that provided by DAB and other digital audio broadcast technologies. FM receivers are largely commoditized and cheap.

System Parameters:
Pilot tone stereo system.
Peak deviation: 75kHz.
Pre-emphasis: 50us.
Pilot 19kHz at 7.5% to 10% of max deviation.
Audio bandwidth 30Hz to 15kHz.

General service planning
ITU-R BS412-7 regards 54dBuV/m as the minimum field strength needed in rural districts and up to 70dBuV/m in large cities. Field strengths can be 6dB lower for monophonic services. These field strengths will result in a weighted signal to noise ratio of >50dB (wtg ITU-R BS468)
Protection ratios
Despite the claims that 'capture effect' will allow for perfect reception in the presence of another signal on the same frequency just slightly weaker than the wanted signal, it won't. Capture effect applies to narrow band monophonic FM only and not broadcast stereo services. In fact, if another station is on exactly the same frequency as the wanted one, it has to be 36dB weaker. Otherwise there will be a worsening of the wanted S/N ratio. However there is an even worse case. For stereo FM, an interfering signal that is 38kHz away from your centre frequency must be 44dB weaker in order not to cause problems. At frequencies further from centre, the situation improves. An interferer that is 100kHz away need only be 6dB weaker and if it is 200kHz away, it may actually be up to 20dB stronger before causing difficulties.

The composite baseband signal

To create FM stereo, the audio has to be first coded into a composite baseband signal prior to frequency modulation. The physical unit with this function may be separate or may be incorporated into an FM transmitter. The sketch below illustrates the composite signal which becomes the overall 'modulating' signal.

FM baseband pic

The baseband extends to as high as 100kHz if all the optional parts are used. In this diagram, the key frequencies are shown and the trapezoidal blocks show the areas occupied by each component. The vertical scales do not represent any particular power level. The audio signals are continuously varying anyway. The stereo coder first sums the left and right audio signals to create the L+R signal and this is all that is needed for mono transmissions. Then the coder subtracts the left from the right and the resultant is amplitude-modulated on to a 38kHz subcarrier which of course creates two sidebands extending now from 23kHz to 53kHz. The 38kHz subcarrier is suppressed, so to enable demodulation of this L-R information, a 19kHz pilot tone is created by exact division of the original 38kHz subcarrier. So, the complete stereo audio baseband now ends at 53kHz. The remaining signals shown are when RDS (radio data system) and SCA (subsidiary carrier authorisations) are used. SCA is used on some transmissions and RDS is also becoming quite common.

This composite signal now modulates the FM carrier. The total deviation must be no more than 75kHz. So, if RDS and/or SCA signals are used, the effective deviation of the stereo audio components are reduced slightly, which does reduce their signal to noise ratio.

Performance of the stereo coder is critical. In order to retain better than 45dB stereo separation, the frequency response of the coder from dc to 53kHz must be within plus or minus 0.07dB and the phase shift be less than plus or minus 0.45 degrees. When RDS and SCA are used, a tight specification must be maintained to the upper edge of whichever SCA is in use, otherwise crosstalk between SCA and main audio channels will worsen. This specification is high but manageable when the coder and modulator are together in one unit but when a composite STL is used between studio and transmitter, the situation is difficult. Broadcasters may be tempted to use composite processing. Don't do it. A composite signal processor alters dynamically the relationship between the baseband components. It is unavoidable. If low distortion and good separation are your goal, do any processing (if you must) prior to the stereo coder.

Bandwidth required for FM stereo

FM is a constant power system. The total power is the same whether modulated or unmodulated. Frequency modulating a carrier results in a large number of sidebands extending out both sides of centre. The power of the actual carrier dynamically changes and can instantaneously become zero with the right combination of modulating frequency and level. However, the power sum of carrier and sidebands is constant -as long as you measure over an infinite bandwidth. So, although the system maximum deviation is said to be 75kHz, in theory you need infinite bandwidth to transmit FM without distortion. This is because the exact relationship of all those sidebands to each other both in amplitude and phase has to be maintained in order to reconstruct the signal perfectly at the receiver. Infinite bandwidth is expensive and not just a little impractical. FM broadcasting internationally resides in the radio spectrum between 88MHz and 108MHz. Even if you had no restriction above the band, you are going to hit zero on the low side, so infinite bandwidth is just not a goer. So, what problems occur if we pass FM stereo through networks with limited bandwidth?

Although there is an infinite family of sidebands extending out each side of the centre frequency, they do tend to reduce in power level as they get further away at a rate depending on the modulation index. So it is just a case of working out at what power level do the sidebands become insignificant enough to ignore. Sounds easy, but it is not. If you are mathematically inclined get a copy of 'Harmonic distortion of frequency-modulated waves by linear networks' by R.G Medhurst. The maths level is extreme and the bottom line is that distortion increases with deviation and with modulating frequency in a complex manner and is caused by both amplitude and phase characteristics of filter networks. Most of us are better off trying to work out the scale of the problem by empirical methods. Fortunately, E.J Anthony of BE (Broadcast Electronics) has done some tests along these lines. He connected a very high quality FM exciter to a precision demodulator via a variable-width cavity resonator. His cavity has only a single-pole characteristic, but nonetheless gives us an idea.

Mr Anthony concludes that good audio performance can be achieved with as little as 800kHz bandwidth. Although that is much less than infinity, it is still generally too large for practical implementations. Why? Bandwidth is usually more restricted than this at the transmission side and always restricted in the receiver. The receiver has to have selectivity in order to reject adjacent channels. That requirement means the receiver filters are much narrower than 800kHz.

Transmission filters

For the transmission side, unless you have only one transmitter connected to the antenna, you need to have RF filters. The filters not only allow multiple transmitters to feed a common antenna without interaction, they are necessary to reduce intermodulation between the transmitters. Regulatory authorities are quite tough on the out of channel emissions from transmitters. Working for BCL (later called Kordia), I was involved in planning some of the transmitting site combiner systems. It was common to have more than 10 FM transmitters of 500 watts to 5kW combined into a transmitting antenna and most channels were spaced 1.6MHz apart. Some, now are only spaced 800kHz from another service. The potential for intermodulation is high and only through solid engineering practices could the issues be managed. The filters at the output of every transmitter were usually part of a tee combiner or constant-impedance (Lorenz) combiner. Filters were either 2-pole for channels spaced a reasonable distance in frequency from a neighbor, but most are now 3-pole. The filter bandwidth is typically 256kHz at the -1dB points.

Why 256kHz. As it turns out, this is what the regulatory authority wants and it is coincident with a rule of thumb long used in the industry called Carsons Rule. This rule states that the occupied spectrum for a FM transmission is equal to 2*(deviation + max modulating frequency). OK, so you know the maximum deviation is 75kHz and the maximum audio frequency is 15kHz. Carsons rule gives you a occupied bandwidth of 180kHz. Sorry, wrong answer! That might work for mono FM, but for stereo the maximum modulating frequency is 53kHz -see diagram above. Now Carsons Rule gives 256kHz, which is what is taken to be gospel. Even that bandwidth is lower than what Carsons Rule would imply if you took the maximum possible baseband SCA frequency of 100kHz into account. Using Carsons Rule then gives you a bandwidth of 350kHz. This tends to be ignored and most broadcasters use 256kHz; the saving grace being that SCA contributes just a small percentage of the overall deviation. The 256kHz wide filters will inevitably cause some distortion and crosstalk- it is just a question of whether it is acceptable. Carsons rule gives a bandwidth that contains 98% of the power but there are many sidebands extending beyond that bandwidth. The work by E.J Anthony outlined above suggests that a 256kHz bandwidth will be much too narrow for acceptable performance, but there is a difference. Mr Anthony used 1-pole filters and we now mostly use 3-pole filters. These are much flatter across the top and continue flat to a wider bandwidth than the 1-pole versions. The down-side being that the group delay of the 3-pole filter will increase at a much faster rate at the filter edges. Will this shape of filter cause more or less degradation of performance than the simple 1-pole case. The answer is I don't know and I have found no real evidence either. I offered to make these measurements on several occasions when I was at BCL but none of the broadcasters were really interested in finding out, or didn't have the green to pay for the work. Some generalities can be made.

The distribution of energy across the sidebands is complex and depends on the instantaneous modulation index.
Mod index M=(deviation/mod frequency).

So, for example if a 1kHz tone resulted in 75kHz deviation, M=75. If a 15kHz tone resulted in 75kHz deviation, M=5. The rate of reduction of the sideband levels is governed by Bessel functions with argument of M. In the first case of 1kHz, a whole family of sidebands extends outwards every 1kHz from the centre frequency. It is only at about the 100th sideband that the amplitude goes below -60dB. So a filter that passed 100kHz each side of centre(or 200kHz wide in total) might not cause much damage to a 1kHz audio tone.
For the 5kHz signal deviating 75kHz, sidebands occur every 5kHz from centre and after the 15th sideband they are more than 60dB down. So, a filter might only be 150kHz wide to avoid problems. The situation becomes complex for stereo signals. Say you fed a stereo 15kHz signal into your modulator with the left channel out of phase with the right channel. Or L=-R in other words. In the FM baseband there is now no signal at 15kHz but there is one at 23kHz and another at 53kHz. The total deviation is still 75kHz; half of that due to each signal, ignoring the pilot. So, for the 23kHz signal, M is 1.63 and beyond the 9th sideband, energy is 60dB down. That therefore needs a bandwidth of 414kHz. The 53kHz signal has M=0.71 and above the 7th sideband the level is more than 60dB down. That needs a bandwidth of 742kHz.
In these latter cases, a 256kHz wide filter would seem to be inadequate, but in fact I don't know whether 60dB down is a good reference level to use and a saving grace is that high audio frequencies in practice are not sufficiently loud to result in maximum deviation, even after being pre-emphasised. Unless thay are not properly limited before coding, that is. Broadcast standards are somewhat more relaxed than what would strictly be called high-fidelity and so the degradations that do occur due to filtering are liable to be within acceptable broadcast limits.

sidebands pic

In Figs 2 and 3 above the vertical scale is intended to be logarithmic. The Fig.2 sidebands are not drawn accurately, but in Fig 3 the relationship is approximately correct. The Fig.3 example is extreme in that a 15kHz audio difference signal is rarely, if ever going to result in full deviation, nonetheless it does illustrate how sidebands can be removed by filtering thus making it difficult to accurately reconstruct the composite signal.

We haven't discussed phase response (group delay) yet and this also affects the sideband relationships. Group delay in a filter starts to increase well before the skirts of the filter are significantly attenuating the amplitude. Group delay in the passband causes distortion and crosstalk due to the influence on the composite signal. The effect becomes worse if filter characteristics are not symmetrical each side of centre.

Delayed reflections

Reflections (ghosts in TV-speak) create audio distortion and crosstalk in FM transmissions. At the transmitter station such reflections are caused by poor return loss antennas. In the coverage area, reflections almost always exist to a lesser or greater degree. It is quite common to hear the excess sibilance (splatter sounds) that are the result of long delay echoes on the radio path between transmitter and receiver. In 1980 an engineer; Mitsuo Ohara of NHK did some empirical tests to determine the scale of this problem. Summarising Mr Ohara, the degree of distortion increases with delay and with audio frequency. For shorter delays, distortion also increases with deviation but that situation changes for very long delays such as exist in the coverage area. At the transmitter site, the echo is caused by some of the power being reflected from the antenna back to the transmitter. The signal then bounces from the transmitter back to the antenna and is re-radiated with a delay equal to twice the propagation time of the feeder. Thus long feeders are most at risk of causing a problem. Of course, long feeders also have more loss so they may well mitigate the problem anyway by attenuating the echo. High powered FM stations have very low loss feeders and most of these will be quite long as well so caution is required in this situation. Mr Ohara determined that crosstalk from SCA into the stereo channels will occur. To meet crosstalk specs of 65dB, an 4us re-radiated ghost must be better than 27dB below the main signal. This is quite demanding because most broadcast antennas do not have this good an return loss. Some mitigation occurs because transmitters will partially absorb the reflected signal from the antenna thus relaxing the requirements. Short feeders are less at risk because the re-radiated ghost is only slightly delayed .

Receiving filters

As mentioned, receivers need selectivity and that is obtained through the use of I.F filters, usually these are centred on 10.7MHz. The common types are ceramic; these are used in all cheap receivers and indeed in many so-called high end receivers. Most of these are not wide enough and also have appalling group delay characteristics; often assymetric. Below are three examples of 10.7MHz I.F filters for receivers.

IF filters pic

Fig.4 is the type more commonly found in receivers. This is not ideal when best performance is desired. Better will be the Fig.5 type, but both these two have a rounded response and moderately high group delay and are typical of ceramic filters. Fig.6 shows a SAW filter which has a flat passband response and a group delay of only 25% of the ceramic versions. This type of filter will allow for the best performance. 10.7MHz SAW filters are not that common and SAW filters can be more frequently found at frequencies of 23.4MHz and 70MHz. Of course these would necessitate a receiver redesign. The use of wider filters in receivers will improve audio performance but may compromise adjacent channel selectivity. Some receivers have two filters; one wide and one narrow, so that the wide one is used in strong signal areas that have no adjacent signals present while the narrow one can be used if interference is an issue, at the expense of higher distortion.

Complying with MED-RSM specifications

This is about the New Zealand situation, but in fact this is not very different from the demands of most regulatory authorities.

Radiated power level: MED issues a spec with the licence. It may have a HRP diagram or define maximum EIRP levels by sector. It will define polarisation. You have to have some means of measuring power either with wattmeter or by using a spectrum analyser attached to a directional probe of known coupling factor. The power of the transmitter has to be set with a knowledge of your antenna gain plus the loss of the feeder.
Frequency Accuracy: Normally must be within 500Hz but LPFM(low power FM) is relaxed to within 5kHz. Check this with a counter attached to a sample of the RF output and remove all modulation for the measurement.
Deviation: Will always be a maximum of 75kHz.
Occupied Spectrum: MED issue a mask or a statement that is in parts usually as follows: (a)Emissions must be more than 25dB down between 182kHz and 240kHz from centre; (b) more than 35dB down between 240kHz and 600kHz from centre and (c) must not exceed -50dBW EIRP at more than 600kHz from centre frequency. How is this achieved? Well assuming your transmitter does not have instability or unintentional emissions, and you do not over-deviate, then you will comply with this specification using only a simple harmonic (low-pass) filter at the output. Most transmitters have these built-in. However, some transmitters are not designed very well and may be unstable and have spurious emissions. This has happened to even commercial broadcasters who eschew professional advice. How to check? You need to borrow or hire a spectrum analyser and preferably someone who knows how to use it. The spectrum analyser will need to be set to a suitable resolution bandwidth; typically 10kHz and also to peak hold for a minute or two. See this diagram for an actual measurement with a mask graticule.

spectrum pic

The transmitter in this case is modulated with a 400Hz tone at 75kHz deviation. Deviation can be checked by noting the distance between peaks which here is indeed 150kHz. This transmitter fits well within the mask to 600kHz each side of centre. Depending on power level, there might be problems outside the 600kHz mask.

Once occupied spectrum parts a and b are checked you need to establish how far down -50dBW is relative to your carrier. If for example, your EIRP is 1 watt, this equals 0dBW, so your spec is -50dBc (dB with respect to carrier). If your EIRP is 10 watts, then your spec is -60dBc and so on. Measure this with no modulation. Then you need to check harmonics at 2f and 3f. They also must be better than the -50dBW EIRP limit. (I have used -50dBW as the spec but yours may be different) It is easy to overload a spectrum analyser when making this measurement and get a result seemingly worse than it really is.

How to check your deviation meter: If you have a deviation meter but are unsure of its accuracy, then you can simply check it while you have the spectrum analyser handy. You will recall from the bit about FM baseband above, that the sidebands and the carrier vary in level depending on the modulation index; M. Well at certain values of M, the carrier (fc) becomes zero and all the power is in the sidebands. The lowest value of M for this to occur is when M=2.405. So, if you were to put an audio signal of 31,185Hz into your composite input (turn off pilot) and slowly increase the input level while watching the spectrum you will see the point where the carrier disappears. That will occur with exactly 75kHz deviation. You will want to set the analyser at the centre frequency and with a span of say 200kHz. If you have no composite input to the exciter and want to use an inband audio signal, you can use the same process. Just feed in 12.5kHz tone, L=R preferably with pilot off, slowly increase level and the point where the carrier disappears will correspond to 30kHz deviation. The span of the analyser will probably need to be only 100kHz for this option.

The ultimate FM system

If your aim was the make the best quality FM system end to end, then the following is my list of what is necessary. Now, this is a little whimsical and I know that this goal is not suitable for commercial broadcasting generally, which is more about quantity than quality, but bear with me a bit longer....

Part One: At the studio

(1) Good quality CD or pc based equipment. Usual problem will be hum or buzz especially in pc's. Sort out your levels.

(2) No processing. Disturbing the dynamics of high quality music is a travesty. Should be outlawed. OK, I know it doesn't matter for pop jukebox or talk type stations but this is about preserving the original music as faithfully as possible. Just last week I heard a familiar Pink Floyd track on a local station. Started quietly as it does and when it came to the loud dynamic bit, nothing changed. The whole thing was flattened to death. I couldn't find the off button fast enough.

All you need is a peak limiter and this is because obtaining the best performance of the FM broadcast system depends on not over-deviating the carrier. Get a good peak limiter; one which pre-emphasises before level detection and then does de-emphasis after the VCA. That is quite separate from the system 50us pre-emphasis you must have in your modulator. Set the limiter up properly for minimal intervention. A good one which was around a few years ago was the EMT (model 266 I think). Made in Germany, some second hand units might be still around. I am sure many clones have been made. This had minimal effect on sound quality and is all the insurance you need.

(3) Studio-Transmitter Link: If your transmitter is not located in the studio building you need a link. These don't come cheap. The best ones now are all digital but is an investment. For the analogue methods, you can either go the composite way, which is quite common in this country or go dual-mono, which uses two radio channels. They each have disadvantages. Composite is obviously cheaper but the end to end link noise performance is not as good as the dual mono link. Some radio stations now operate single-frequency networks in certain areas. These always need digital linking. But, the best quality option -no link at all.

The transmitter side

(1) Stereo coder: Again these days all-digital ones are around. It is important to get this right; the key to quality FM is the composite signal. If you are using SCA or RDS pay attention to the relative levels. Using SCA and/or RDS does reduce the deviation for the actual main audio and compromises the signal/noise ratio a little but of most concern is the potential for crosstalk from SCA into the stereo channels which sounds like low level monkey-chatter. Ideally avoid SCA and RDS.

(2) Exciter and transmitter: Exciters which use DDS (direct digital synthesis) modulators are around now and these offer the most precise frequency modulation. However there is no reason why analogue modulators cannot be very good also. Dodgy modulators can be unstable, drift and result in spurious emissions, which will force you to try and filter them out.

Rule No.1 is don't overmodulate. Overmodulation has severe effects. First, you will almost certainly be in breach of the regulatory authorities in terms of occupied spectrum. Second, your modulator will almost certainly be distorting. Third, if there are RF filters after your transmitter, they will result in further distortion and crosstalk because you are creating larger sidebands and making them extend further out from carrier. Fourth. You can splatter adjacent frequency licences. Fifth, receivers cannot cope with excess deviation and will distort, especially bad will be distortion of sibilant vocal sounds. Yes, I know it is tempting to push things. More deviation means more volume. You imagine that people are routinely scanning all the FM broadcasts and will be magnetically attracted to the loudest one and then make you their No.1 station. Absolute nonsense! It is the content of the station and sometimes the broadcasting personalities that are the reasons for preferring one station over another. I have heard stations so badly overdeviated the distortion from even a good FM receiver was intolerable; the studio operators didn't realise because they were listening only to the studio output.

Power of transmitter: Enough to create planning guideline field strengths in the desired coverage area. 1 watt EIRP (stereo) will reach 11km in rural districts but only 2-5km in urban areas. Assumes line-of-sight of course. Want to double the distance; you need 4x the power. Transmitter power does NOT determine the loudness of your station.

(3) RF filters: Avoid if possible. If you are on a multi-user site feeding a common antenna then filters will be mandatory. These may form part of the channel combiner system supplied by your transmission partner. Usually in this case, the filters will have the generally accepted performance even if this does prevent the ultimate in performance. You should confirm the specs for this though and get a plot of the filter response and group delay. Ensure it meets bandwidth requirements and make especially sure it is symmetrically tuned. . If you are doing it yourself and want a bandpass filter for security, there is nothing to stop you engineering a wider than the usual 256kHz filter.

(4) Antenna: These can be either simple dipoles for low power broadcasting, or quite involved custom arrays for the big broadcasters. The bandwidth of these is not an issue but the return loss might be. Aim for at least 20dB. For a low power station, to avoid antenna reflection performance issues, you could use a lossy feeder or even an RF attenuator. Transmitter power is cheap and easy when talking just a few watts. If you need to put 1 watt into your antenna, you could use a feeder with 3dB loss and then put in 2 watts. Achieves the same EIRP but will mitigate feeder ghosts by 6dB. Not very practical at more than say 50 to 100 watts though. At these sort of powers a circulator at the transmitter output will solve all the return loss and intermodulation issues. Above about 300 watts, circulators are massively expensive, although in a paralled P.A transmitter, circulators could be placed at each module output before the module combiner.

The receiving side

First, the antenna. An outdoor, directional yagi type antenna will be the best. I know a bit of wire hanging out the back of the receiver does work, but it will never give the best possible performance. You need a good antenna not because the signal might be weak, but even in strong signal areas, you can reduce the pickup of reflected signals (ghosts) and avoid getting that nasty excess sibilance. This is more of a problem than you might imagine and occurs particularly in hilly areas. High quality stations are more affected and that's you, right.

Receiver. There are many choices out there, but from our earlier discussions you want one with sufficiently wide filters. Published specs are intended to mislead in many cases, but you have to start somewhere. Price might not be a guarantee of good performance, however, paying too little is a guarantee that performance won't be good. Receivers that mention pulse-counting discriminators probably have an edge and you could look at multiple bandwidth options, indicating that the receiver manufacturer has an understanding of the influence of I.F bandwidth.

Final remarks

Well that's it. Do all the above things and a top performance FM system is yours. It isn't possible to call it CD quality, but the only thing lacking is some dynamic range. i.e the signal to noise ratio cannot be as good as CD. It is on a par with vinyl though. The slight limitation of top end frequency response from 20kHz on CD to 15kHz on FM is going to be noticed by only a few; perhaps 1% of the population and then only when critically listening. So don't accept that this is a major issue. You can start fixing up your existing system bit by bit and that is the advantage of analogue - improve something incrementally and get an incremental improvement.

BACK to the TOP

Sources used:
ITU-R BS450-3 Transmission Standards for FM sound broadcasting at VHF
ITU-R BS412-7 Planning Standards for FM sound broadcasting at VHF
ITU-R rec 641 Determination of radio frequency protection ratios for FM sound broadcasting
Harmonic Distortion of frequency modulated waves by linear networks - R.G Medhurst
BE: Optimum bandwidth for FM transmission Edward J Anthony
BE: The composite signal-key to quality FM broadcasting Geoffrey N Mendenhall
Mitsuo Ohara NHK IEEE Transactions on broadcasting, Vol BC-26, No.3, September 1980.

Axino-tech Consulting & Services , August 2010.

Comment - Audio compression useful in vehicles

From: Axino 16 May 2012

In the article I was critical of the use of audio processing, aka dynamic range compression. I have had a couple of people comment to the effect that compression has its uses and is necessary for listening to the radio in vehicles.

Certainly it is true that the noise levels in vehicles will mask quieter sounds and compressing the dynamic range can assist. This is particularly true for the talk-style stations. However, my view is that music stations at least should not be 'crippling' the music by using compression just so that it sounds better in cars.

The better approach is for the receiver end to have the choice to compress, or not, rather like the options in modern digital TV's and set-top boxes. They provide for a 'volume limit' or 'night-mode' setting which intentionally restricts dynamic range, mainly to try to even out levels from ads. If the mode is set to 'normal' then the full dynamic range is available. I know these options work better in a digital environment, which FM radio is not, however, car radio manufacturers could still implement the option quite easily.

I don't know if such a facility exists on any particular car receiver model, so if anyone knows they could let me know.

Back to the TOP