Output from a "canned" Oscillator
    Of course, the output is at a TTL level: a squar(ish) wave between zero and five volts. But most TTL outputs are not at all linear. This means that available drive current varies, depending on the output voltage. Standard TTL outputs are pretty good at pulling current toward ground (at the "zero" volt level), but they're none too good at outputting current (at the "five-volt" level). Unfortunately, a power amplifying transistor needs significant current coming out of the driver, and into its base.  A PNP final amp would work better than an NPN transistor, but NPN finals are much more common.
    You could add a tuned circuit between the oscillator output and base: with appropriate impedance matching, you could get enough base drive. This is the solution that the N.J. "fireball" rig uses. A five-element PI filter is used between the TTL output of the oscillator and the base of the amplifying transistor.
    Another approach is to boost the TTL drive available with a bus-buffer TTL chip. These have more current output, and more significantly, can drive much more current into the base of the final amp.
Adding a few in parallel helps provide even more drive. By using a tri-state buffer, the output drive can be turned on and off with a TTL-compatible keying signal (or the collector of a NPN keying transistor, with its emitter grounded).
    With this approach, a simpler base-drive circuit works fine, and no tuned circuits are needed, except after the final amp.
    The 74F125 chip has four buffers. The remaining unused buffer could be employed as a TTL oscillator, if you happen to have a bare crystal rather than a canned oscillator.

Choice of buffer chips
    There are a LOT of suitable buffers that'll do the job. I've chosen one that was cheap and available. It is available in a number of different logic families, including standard TTL, low-power shottkey TTL (LS), advanced schottkey TTL (AS), Fast TTL (F), High-speed CMOS (HC), and Advanced CMOS (AC). At 28MHz, we require that the buffer switch between zero and five volts quickly. And we need the high output current too.
Chip        OK?        why?
74125        no            too slow, not enough drive current
74LS125    no            not enough drive current
74ALS125    no        fast enough, but marginal drive current
74HC125    OK        could be better if faster: can run up to 7v supply.
74AC125    good        very fast, low output resistance (lots of drive)
74AS125    good        very fast, pretty good output drive
74F125        good        very fast, pretty good output drive

Output Amplifier
    You could use a 2N2222A, but its a bit skimpy on heat dissipation. A TO-39 transistor like 2N3553, or 2N3866 is safer. To get reasonable power out, its collector should be biased at more that 5v. A 12volt supply is almost standard. But all the TTL circuitry requires something close to 5v. A 78L05 regulator will work fine, but the arrangement shown is cheaper. A simple NPN plastic-case transistor is biased to drop the 12v supply down for all the TTL circuits. If your 12v supply is adjustable, then tweak its voltage so that the TTL gets close to +5v on their supply pins.
    The base-drive circuit for the final amp was tried with a 74LS125 and a 74F125. With the 74LS125, output power was about 200mW. With the 74F125, output power increased to 1.6W. Some of the resistor values could be tweaked in order to drive the final amp at a 50% duty cycle, or a bit less. If its overdriven (past 50% duty cycle), efficiency suffers. With 74F125, total transmitter efficiency is about 50% - that's about all you can expect.
    The output five-element PI filter uses air-core self-supporting coils, mounted above a printed circuit board. These were pre-tuned by tacking a 100pf capacitor in parallel, and finding resonance with a grid-dip oscillator. All capacitors are silver-mica. The filter was designed for 35 ohms input Z, 50 ohms output Z. It has a chebychev response with 0.1dB ripple. Cutoff frequency was a little higher than 29MHz to make sure that capacitor tolerances will allow 28MHz to pass with no loss.
May I recommend Bob Lombardi's FDS2 program for designing these type filters.  It is available via FTP from the QRP-L archives.
FDS2 5-elemet PI filter values:
C4    177pf
L1    259nH
C5    259pf
L2    370nH
C6    124pf
    I started with Bob's AIRWNDL program to find how to wind L1 and L2. After checking with a grid-dip oscillator, their inductance was a bit high. This is likely because when winding onto a convenient 1/4" shaft of a variable resistor, the coil diameter increased while the coil "relaxed" after winding. A turn or two was removed, leaving L1 at about 7.5 turns, and L2 at 10 turns. These coils were not stretched: there's no air space between turns.
    RFC2 is a commercial molded choke. RFC1 was lifted from a PC plug-in card . It had filtered a serial-port output line. A ferrite bead with three turns should work well too.
    A 120 pf capacitor could be used for C6, but I used a 47pf in parallel with a 75pf (both silver mica). You can make up many non-standard values with two smaller caps in parallel. The RF current will divide between the two caps, so if they're lossy, parallel caps will do better than a single lossy capacitor.
    A small finned headsink was added to Q1, since it does get warm.