RDS Encoder Stereo FM Motion + good benefits.

Hi all, the proposal later I will do is the forum for those who are interested in

participating in the design and prototyping of one (or more) systems for stereo and

RDS FM encoding. The objective is to design the first part of the stereo encoder and

once we have a design that meets the claims, incorporate a RDS. I'm sure that some

of the blocks of the encoder can also be used for stereo RDS encoder therefore save

resources when enconders implement both simultaneously. In advance notice that

the post is quite long, but it sure some going find interesting ... There will be awards

for those who read to the end, hehehehe. I've been mulling over this issue for some

time as the systems that have been implemented always have some weakness: They

are either expensive (DACs very high speed and CPUs), or are not broadcast in

broadcast quality (such as those using the monolithic type BH1417 IC, BA1404, etc)

or very unstable and difficult to adjust (MC1496), or do not work well in all FM

receivers (fault Precision 38kHz subcarrier). Of those who I liked more was to Pira.cz (

http://pira.cz/eng/stk2en.htm ), but there are things I do not like it. What I dislike is

the deviation of the 38kHz carrier a few tens of hertz for not using a crystal that is an

exact multiple of that frequency, which may for most recipients to work reasonably

well, but for some decoding fails especially in cheap rate SDR receivers (some only

synchronize the phase of the pilot, but do not adjust the frequency but synthesize

directly using 38.000kHz carrier FFT), which are increasingly more common to find in

mobile phones, devices mp3 / mp4, etc. I also think that the Pira wastes resources

but gets good performance. I think it's possible to do something of the same quality

with simpler elements (eg a cheaper MCU) from those based on direct multiplexed

using a square 38 kHz and two type switches 4066 hardly speak and suffering of all

evil, much background noise due to the number of harmonics in the switching leak

can not finish being very close to the passband and a very poor channel separation,

about a 15-20dB due to lack of compensation levels and lack of adjustment of the

phase of the pilot, once senoidificada, with respect to the subcarrier. So the first goal

is to make an economical encoder, as accurate as possible (at most 10Hz / 20Hz of

deviation from the carrier 38kHz, and if you are under so much the better, as above

20Hz deviation in some host family cacophonous effect those with SSB receivers

occurs). Desirable to have good SNR margin of 60dB if possible, and channel spacing

according to the SNR. If you have a SNR margin of 60dB is not worth getting a

separation of greater than 60dB channel because everything that falls below that

level will be masked by background noise. This necessitates the use of a quartz

crystal that is an integer multiple 38 kHz carrier, otherwise we would be forced to use

a PLL / Prescaler / Divider ratio of frequencies, etc., and that much more expensive

project. Also formerly was difficult to get accurate crystals but today with the Internet

and the fact that cheap systems also used, it is not that hard to get them. In fact this

is a design project that was parked in a while and has revived thanks to that

happened a few weeks ago I came across an offer of 4.864MHz crystals very

interesting and I ended up buying. With them I've started doing my first tests. These

crystals so I understand are very hard to get but give the advantage of providing a

multiple of 38kHz which is exactly a power of 2, ie 128 * 38kHz = 4.864MHz. This

means that dividing by 2 ^ 7 crystal frequency we obtain the frequency of the

carrier, and dividing by 2 ^ 8 have the pilot frequency. 've also got 4.332MHz

crystals. This frequency is not a multiple containing only powers of two, but there is

something very interesting ... It contains a power of 3 to 19kHz, allowing it to achieve

57kHz carrier for RDS addition of 38kHz and 19kHz pilot. Summing , with the glass of

4.864MHz is easier to design a stereo encoder construction but slightly complicates

RDS (not being a multiple glass 3). However with the encoder of 4.332MHz slightly

complicated but the carrier stereo RDS is easily obtained. I'm also waiting for me to

arrive and 38kHz crystals 7.600MHz. These crystals although with certain drawbacks,

are also integer multiples of 38 kHz (1 * 38 = 38 200 * 38 = 7600) and using some

ideas can implement higher quality encoders used for both (these crystals are quite

common in find when searching a bit online, as they are used for encoding BA1404

monolithic (38kHz) and BH1417 (7.6MHz)). Summarizing. The designs should be

directed to use one of these four crystals (if you find another possible crystal of a

multiple frequency of 38kHz and accessible to get speak ye because I have not found

it): 7.600MHz (very accessible and cheap, stereo nontrivial RDS is not trivial).

4.864MHz (Rare, stereo trivial, nontrivial RDS). 4.332MHz (Less rare, nontrivial stereo

RDS trivial.) 38kHz (Relatively affordable, not expensive, not trivial stereo RDS

nontrivial) Well, I've started experimenting with that of 4.864MHz but have also

generated a stable carrier using 38.000KHz of 4.332MHz. When I reach the other

crystals I will test them. We start with the theory: For those that have not yet

clear how mathematically build a MPX signal to FM from two audio channels, I will not

extend much since these links explains it better than I can explain ...

http://es.wikipedia.org/wiki/FM_estéreo http://transmitters.tripod.com/stereo.htm But

I will summarize all the above theory in the usual formula for the multiplex signal:

MPX = (Vi + Vd) + (Vi-Vd) * S (38kHz) + 0.1 * P (19kHz) where S is the sine of 38 kHz

subcarrier and P 19kHz pilot signal. Vi is the audio of left channel and Vd is the audio

signal from the right channel. The pilot signal is not involved in the coding process

but adds 10% to be used to synchronize the stereo subcarrier. So for simplicity we

put the MPX signal as MPX = (Vi + Vd) + (Vi-Vd) * S = A + B * S At the receiver is

demodulated B = (Vi-Vd) and filtered A = ( vi + Vd) thus obtained two channels

separately, one sum (A) and a subtraction (B). In fact I saw channel is recovered by

summing both signals (Vd is canceled), winning 2 * Vi and Vd channel recovers

subtracting both signals (Vi canceled), winning 2 * Vd. Of how balanced levels are A

= (Vi + Vd) with respect to B = (Vi-Vd), both in amplitude and phase depend on the

channel spacing. This means that if for example I saw of A is not exactly the same

that I saw in B, the receiver can not be canceled altogether Vi subtracting B from A

and therefore the channel you a part of the channel appears Vi and vice versa, so

that the channel spacing will suffer. Hence, for a good stereo encoder need three

things: -To the LV and RV of signal levels are the same as the LV and RV signal levels

B. -Make 19kHz pilot signal is exactly in phase with the 38 kHz. That the above

conditions do not vary with operating conditions, temperature, humidity, etc. By an

almost analog encoder, using operational to generate the signals amount A (Vl + Vd)

and subtracting B (Vi-Vd), and then use a balanced modulator to obtain a

multiplication in four quadrants (ie both negative voltages as positive) as the MC1496

to generate the B * S signal, achieved a very high separation. Unfortunately the

stability conditions are short and that by varying the values of the components with

the operating system then undergoes channel spacing, forcing readjust operating

conditions. The MPX signal from the above formula we can write equivalent form of it

this way: MPX = Vi * (1 + S) + Vd * (1 - S) The transition from one formula to another

is mathematically simple and you just have to operate with a little algebra and

remember properties of addition and multiplication. It seems that now the shape of

the signal generated is complicated by two multiplications appear instead of one, but

what is between the parentheses, by accounting for 1, and the S signal is a sine wave

(the signal S varies between 1 and -1), never be worth less than 0, in fact worth

between 0 and 2. This causes the two-quadrant multipliers are instead of four, which

is easier to achieve. There is a time when MPX is 2 * Vi + 0 = 2 * Vd * Vi and another

time, 180 degrees out of phase, or what is the same for 38kHz, about 13.16 us later,

MPX is 0 * Vi + 2 . * Vd = 2 * Vd believe the MPX signal is composed by two sub-

signals, MPXd and MPXi, that would: MPXi = Vi * (1 + S) MPXd = Vd * (1-S) and then

MPX = MPXi + MPXd + 0.1 * Driver. This other way of expressing the MPX signal

suggests that both the channel I watched as you are slowly alternating in time.

However at a certain intermediate point not only has Vi or Vd but a combination of

them. Therefore we have used semidigitales techniques to create multiplications (1 +

S) and (1-S) for Vi and Vd with achieve an approximation of MPX. I say approximate

because these techniques try to get an S that looks like a sine greater or lesser

extent, but there will always be some difference between digital signal created by the

encoder and the pure sinusoidal signal S. The approach to economic modulator is the

use of a C instead of the signal S which rather than sine is approximated by a square

of between +1 and -1. In fact the signal (1 + C) is a signal that is 2 0ºC-180 between

180 degrees and 0-360. The signal (1-C) signal is an offset 180 ° from the above, so

that it 0ºC-180 0 2 between 180 and-360. Since MPXi MPXd and signals are

generated in the same way, with the carrier 38kHz outdated 180, we can consider

only the aspects of the generation of the MPXi signal (carrier of the left channel) to

be recícprocos in the generation of the carrier of the right channel, or MPXd. The

multiplication Vi * (1 + C) easily achieved by making 0ºC-180 this value is 2 * Vi, and

between 180 and 360 this value is 0. With You reciprocal channel, between 0 ° and

180 ° is between 180 0-360 will be 2 * Vd. Without But when we get MPXi = Vi * (1 +

C), since C is a square wave of amplitude 1, this is composed of a series of odd

harmonic fourier sine infinite S, which are as follows: MPXi = Vi * (1 + 4 / PI * (S + 1/3

* S (3) + 1/5 * S (5) ...) This is something that is not particularly a lot like the original

form MPXi = Vi * (1 + S). Where S (3), S (5), ... are the odd harmonics of the

fundamental carrier comprising the square. If not completely eliminate these

harmonics in the receiver a phenomenon of aliasing occurs causing appear ghost

tones and artifacts in the audible signal that should not be there, so the SNR is

greatly suffers and the audio quality is quite degraded. To do this, you need a filter

that does not distort the phase of the passband and is abrupt to remove from the

third harmonic, or in other words, they allow up to 53kHz without altering the phase,

but annulled the maximum frequency of 95kHz on. This is very difficult to do so

stereo encoders that use a square to generate signals MPXi MPXd and have to choose

between a poor SNR or poor separation, depending on whether filtered with a filter

that distorts abrupt phase or a less steep filter phase distortion less but allow some

of the harmonics. To complicate matters further, imagine that through an ideal filter

that cuts to 53kHz, eliminate harmonics S (3), S (5), ... In Then we get a signal MPXi

= Vi * (1 + S * 4 / PI) What should range between parentheses sinusoidally between 0

and 2, however as S worth more than 4 / PI and at least -4 / IP, the parenthesis just

being worth a maximum value of 2.273 and a minimum value of -0273 ... I mean, the

moment when I saw the sign should be worth 0, it -0273. This implies that the

composite signal, the MPX moment that you should be present only, ie, 0 + 2 * * I

saw you there is actually this: -0273 * 2273 * Vd + Vi. This means that 12% of the

channel interferes saw you channel and therefore we can achieve maximum

separation eliminating all harmonic is -20Log (0.12) = 18,41dB. Poor separation. In

order to improve this separation, anyway worth far MPXi at its maximum value

(because its value can vary with gain control, we have to make its minimum value is

0. This is the full signal adds this: MPX (final) = MPX (filtered) + (4 / PI-1) * (Vi + Vd)

=> MPXi = Vi * (4 / PI + S * 4 / PI) = Vi * (1 + S ) * 4 / PI Now although the factor 4 /

PI, it is important that the signal varies, ie, what is between the parentheses is

between 0 and 2 so MPXi worth between 0 and Vi * 2 * 4 / PI. This means that if we

implement a multiplexer that you stagger signals and Vi, the output of the

multiplexer should add a 27'3% of the sum signal (Vd + Vi) and thus increase ..

considerably channel separation Sadly almost no implementation of this type that

uses 4066 type switches adds this offset is shown here graphically what has been

explained: In Figure 1, the Square is the signal that is used to multiply Vi, and the

theoretical signal S which should vary between 0 and 2. In Figure 2, you have the

original square wave and sine wave REAL resulting after filtering the third and

successive harmonics. Note the negative peak of the sine. Figure 3 shows the square

wave offset adding a level of 0.273 * Vi, and the resulting sine wave in any case just

to be below zero shows. But even if we can make the signal levels, we have the

problem of finding a filter completely remove the third and successive harmonics and

leave intact the phase of the passband. So what replaces the sine S in the equation,

can not be a square plain but it has to be something else. In my designs I will use

several ideas to synthesize the S signal with a minimum quality that allows using a

less steep filter but respecting the phase, while the compensation, if necessary, is

less than in the case of the square wave. The most logical thing is that we do not

need to remove the harmonic S (3) because the wave we use harmonic that does not

exist. We may use a system to cancel the third harmonic signal we use to mimic S.

Say you have a square wave C we have used before. Now we go through a delay that

departure C shows the same signal but with a lag of 60 °. The third harmonic will

have a phase shift of 60 * 3 = 180 °, ie have -S (3). If we now add this wave with the

original wave, the third harmonic of the resulting wave will be S (3) + (- S (3)) = 0.

That same wave divide it by 2 to restore the original levels. Thus we have an M-wave

(modified sine wave or modified square) to remind the original square, but lacking

the third harmonic. Furthermore, the process can be repeated again first by moving

36th and it then by 25,71º to cancel the harmonic 5 and 7. The harmonic annulled 9

to cancel the third harmonic and the harmonic 9 is a multiple of 3. A square that has

been Delayed 3 times and was averaged as many times ultimately a wave that

contains only 9 levels (enough to be generated with a DAC of 4 bits) and the first

harmonic is number 11 (ie, at 418kHz, and far from we are interested in the band and

therefore easily filtered without altering the phase of the passband). Although the

signal only has 9 levels once completely filtered the remaining harmonics, we have a

signal of 4-bit resolution. We have a pure sine wave (which must be compensated as

long as remaining harmonics have to offset some but far less than the original

square). The process of peeling away harmonic averaging the same square with itself

lagged shown in the following image and repeated several times for different

phases ... In fact this system of harmonic elimination of interference is known as

comb filter (as it not only eliminates harmonic if all its multiples, showing that the

filter response is apparently a comb) and works with any input waveform. A modified

example includes eliminating harmonic waves where instead of a square have a

triangular, and parabolic. But take again modified wave from square ( it is easier to

multiply the triangular or square dish since only two states). Now suppose we

implement a square 38 kHz and 60 desfasamos. These two signals are used to

multiply Vi so that we generate the following expression: MPXi = Vi * (1 + M) where

M is now 2 * sqrt (3) / PI * (S + 1/5 * S (5) + 1/7 * S (7) ...). We now see that no longer

exists S (3) and harmonics start at S (5) which is 38 * 5 = 190kHz. It is now much

easier to filter out these harmonics, we can use a filter that is not as sharp and better

conserve phase. If now we filter, we have to factor 2 * sqrt we have (3) / PI = 1.103:

MPXi = Vi * (1 + S * 1.103), ie, that uncompensated have reduced crosstalk from

12% to 4 , 9% giving 26,3dB separation. It is not a great achievement, but more than

8 dB square wave. Also just have to compensate with 10.3% of the sum signal

instead of 27.3% of the case of the square wave. In the picture below are the figures

before but now applied to a modified waveform to eliminate the third harmonic I

Ideally to implement multiple filters with various offsets to cancel up to harmonic 7 as

shown in the pictures above, but with analog phase shifters easier the more you go

any phase so just having again harmonic levels does not filter the output filter.

However we can use a digital oversampling to generate the exact sine wave without

harmonics to number 11. This requires that the count frequency is a multiple of such

harmonics to the divide by those amounts frequency is accurate, that is, the

frequency original has to be a multiple of 3, 5 and 7, making x105 be asserted.

Unfortunately we can not get the necessary to achieve oversampling x105, since the

value of the glass should be 3.990Mhz glass. Someone might think you could use a

4MHz crystal and holy Easter, with almost almost gets to have a 4MHz but by dividing

by 105, the resulting frequency would 38,095kHz far from our intention to get the

exact 38,000kHz. However if you just want to eliminate "digitally" the third harmonic

and we have a decent filter to eliminate the fifth and succeeding, we can do it

relatively easily using a glass 4.332kHz (since it is a multiple of 3), readily available

and step and get the 57kHz signal to feed an RDS enconder. Here I will propose two

systems for achieving encoded using a modified waveform to elimnar the third

harmonic. The first generates a square 38kHz using a crystal of that frequency, and a

phase shift network 60 generates an offset signal (the phase shifter may be an RC

network which delays the square for 4.4 us and a schmitt trigger to re-shape the

square signal. So we used to combine two signals in a summing obtaining direct

multiplication of both channels signals (1 + M) and (1-M). The remaining blocks do

not need much explanation. Is the compensation network to cancel crosstalk, and the

system to generate the 19kHz pilot signal using a bistable type T and sine wave

shaper (either a 19kHz filter with a very high Q), plus another delay to adjust the sine

phase to minimize again the crosstalk. The second system does the same as the first

but the phase signals are generated by a sequential counter 6 states. This system

has the advantage that by designing the gap 60 may not be varied by the operating

conditions, cold, heat, humidity, etc. Otherwise is the same as the first, although it

uses a crystal of 4.332MHz and allows also generate the 57kHz RDS carrier. Another

method for multiplexing with a sine: In addition to the cancellation of

harmonics, it is possible to synthesize the signals (1+ S) and (1-S) by other methods.

The encoders used in oversampling DACs is the use of two-quadrant multiplier or a

DAC multiplexing at different levels. The typical DAC is one that makes use of an

array of resistors to provide at its output proportional to the analog input digital

signal level. The best known of this type is the R-2R ladder networks, and weighted

resistance (when trying to generate known periodic signals). Other DACs lesser

known but more common because they work almost entirely in the digital domain,

are DACs with pulse density modulation, or PDM. Not to be confused with PWM or

pulse width modulation, which is another way to implement a DAC but not as digital

modulation ... The PDM is also known for its construction, as Delta-Sigma modulation

(or Sigma-Delta). A Delta-Sigma modulator does is track analog pulse signal

generating more "together" the higher the level of the analog signal. For a more

detailed description please visit the wikipedia link on this modulation:

http://es.wikipedia.org/wiki/Modulación_Sigma-Delta A feature of this system is to

achieve a good resolution is sufficient with little bits whenever we can modulator

operate much faster than the frequency we want to synthesize. In fact it is quite

common to see Delta-Sigma modulators using only one bit. The characteristic of this

modulator is based on the harmonics instead of going equally moving along the

spectrum, concentrate on the top of it, so a simple low-pass filter with frequency cut

away the area where the density is higher harmonics, faithfully reconstructs the

original signal. We do not need ADC part, ie, the crawler generates bits. We can

generate a table containing bit Delta-Sigma sinusoidal sequence at a certain

frequency, and storing the table in a fast memory, either an EPROM, a RAM, or a

microcontroller. However the DAC modulator hand, can be as simple as an RC filter

with the cutoff frequency two decades below the sampling frequency (40dB

attenuation getting annoying harmonics). That is, a simple switch switching quickly

put at 128 times the frequency you want to synthesize, and then the filter we obtain

the corresponding audio channel multiplied with the sine (1 + S) with a fairly

acceptable resolution. The image below shows the waves when the frequency of the

bit stream is 128 times the subcarrier frequency, ie 128 * 38 = 4.684MHz. We can

also easily generate several waves while since we have several bits in memory for

each temporal position in the sequence. For example we can use one or two bits both

to generate waves of 38kHz and 19kHz pilot. The following schemes are based on

this principle. The first uses an EPROM type memory and a crystal that powers the

clock counter that said memory addresses. To generate the 19kHz signal must be

repeated 2 times 38 kHz, so a 4.864MHz crystal 256Bytes needed, with one of

7.6MHz, 400Bytes, and that of 4.332MHz, 228Bytes. Except 4.864MHz crystal, other

crystals for the next to last position of the table causes the counter is reset and

return to the beginning of memory. Although not shown in these diagrams,

eliminating much of harmonics, the level of the fundamental wave is greatly reduced,

but not the DC level. This means it is necessary to compensate but in this case

instead of adding to the MPX (Vi + Vd) signal subtract a percentage of it, equivalent

to the level of the signal over the final composite signal. Testing breadboard for

try to generate both 38kHz and 19kHz pulses as sinusoidal waves corresponding to

these frequencies, the moment I tried using the table of Delta-Sigma have generated

waves to construct the graph seen above. I used two bits to encode the 38 kHz signal

of 2 bits PDM, one extra bit to encode the same signal with a resolution of 1 bit, and

3 bits over 19kHz to generate two bits and 1 bit. In total I have used 6 of the 8 bits in

a parallel EEPROM Xicor X2804 model (that is what we had for the disaster box). How

many bytes are not've programmed manually. I used a crystal 4.864MHz with a

74HC04 gate to generate the wave train feeding the meter. Precisely the problem

with my installation and that has limited me quite the results have been used a low-

speed version of the 4040, such as the MC14040 works well up to 2MHz to 5V (I gave

a pulse train more than twice as fast). The counter had, but Q1, ie A0 in memory,

never down to 2.5V so you'll always remained to 1. The result is that with this I could

only counter directional odd memory addresses at that speed so that oversampling is

reduced from x64 to x128. Besides the lack of speed makes addresses crawl in the

series (they lag behind the pulse train) which makes the resulting wave is not

perfectly sinusoidal but trailing slightly behind. I hope to correct these problems

when they get a 74HC4040 which is 10 times faster than the MC14040. So far these

are the images that have been gathered from the tests. Here I show an image

generating portion of the Delta-Sigma mounted on breadboard: Here is a picture of

the square wave generated by the glass: In the next photo, output Q8 meter (19,000

kHz ie 4864/256) In the next photo The Q7 output (38,000 kHz is 4864/128) In the

next photo, you have the PDM signal 19,000KHz a bit with the clock of 19KHz. In the

next photo, you have the PDM signal 38,000KHz of one bit with the clock of 38KHz. In

the next photo, I placed an RC lowpass with a 10k resistor and 470pF capacitor at the

output of each of the PDM signals of 19 and 38 kHz. 19kHz signal is reconstructed

fairly well despite being only 1 bit, but the asymmetry is due to the drag bit that I

mentioned before because of the counter. However although it should also become

good, PDM 38 kHz signal appears formed like a triangle. Already in the last photo, the

drag bit for PDM 19kHz signal and two bits of resolution did not put as although

improved symmetry, an area had a strong distortion. However the 38 kHz if it

improved by using 2 bits (basically place 2 10k resistors at the two outputs and

connected to the same 470pF capacitor integrator making and summing the two

signals). Now at 38 kHz is clearly synthesized sine. So far these are the tests I've

done. I want to get more integrated logic, as more counters, flip-flops, faster gates,

etc. So I can deploy multiple versions of the designs I've shown and see what

everyone is more stable, less noise, more separation, it is friendly to the RDS, etc.

Hence the title of this POST. I am a person of limited means, and I also like to make

this project and remain as collaborative design forum (then we each do what he

wanted with it). Proposal:

So I wonder if there are people willing to test the ideas explained here ultimately help

achieve some of the designs I have proposed either materializing or adding more

functional blocks (pre-emphasis filter, etc). And once we Stereo few modulators,

design the RDS modulator that best fits the stereo modulator (using frequency

master clock for example). It is not a project that has to be done quickly, that is, that

I do not care if it costs a year or more to complete, but if you get something decent

we could have a design that became popular and thus give some publicity to the

forum. As it is cheaper to buy many components when shopping on ebay and other

online sites have more crystals . those who will use would have to share the following

crystals are difficult to obtain in principle to share one glass per person: . 10

4.864MHz crystal 4.332MHz crystal 10. And as soon as I arrive in the mail in addition I

will share the following : . 10 crystals for 7.600MHz . 10 crystals of 38kHz not charge

anything for the glass, all I would ask that the shipping costs for Spain would be

approximately 1 euro, and would be done by mail. For outside Spain especially Latin

America shipping is multiplied so that initially ideally involving Spanish foreros

though someone get the crystals on your own or is willing to pay the cost of

international shipping, so be it . Also, if anyone needs more than one glass should

justify why you want more than one. It must also be someone on the forum, ie,

someone who actively participate. Although it has many messages, you must have at

least relevant messages or odd contribution. Those interested in a crystal that

comment that they want to participate in the design in this thread. The design should

begin to realize or implement, or at least parts of it, within 6 months from the date

you mail the glass. No matter then later to finish the project, it is important to at

least start. Well, I hope your ideas and contributions to the project, and of course,

your requests. Ahh, if someone wants to implement the project using a generated

table in RAM high speed my disposal 10 static RAM chips HM6116 recycled arcade. I

could send one of them along the glass (provided the total does not exceed about 20

grams). Regards, and thanks for get to read to the end.