These numbers may not be what one would expect, thinking that doubling the resistance cuts the current in half and therefore the power. Looking again, 5A@20V produces 100 watts into 4 ohms, the same 20V produces 2.5A into 8 ohms for 50 watts. Doubling the impedance cuts both current and power in half at the same voltage. At the same output power, increasing the impedance increases voltage but
decreases current . For purposes of this AppNote, "rms power" measurements are taken at the point where either clipping begins to occur or thermal protection begins to engage, whichever is smaller. Input signal used is a sine wave,
constant amplitude driving the amplifier to the desired measurement point.
HEATSINKING
When it comes to heatsinking, bigger is better. Too small a heat sink will result in premature thermal shutdowns, particularly when more than one IC is attached to the same heatsink. Production economics necessitate going with the minimum sink that will do the job, do-it-yourself builders should not scrimp. One of our favorites is PEI682 extrusion, which carries a 2.3 degree C/W rating for a 3" length. Five inches does nicely for two ICs, though we have an experimental unit that has 8 of LM3886 on an 18" length. Especially in multi-chip situations (parallel, bridged or parallel-bridged) heatsinking should be sufficient to
ensure that thermal protection is engaged from the chip's limitations in transferring heat from the die to the heatsink rather than from the heatsink to the ambient air. The big advantage that comes from operating LM3886s in parallel
is not so much having twice the output current available as it is having twice the heat dissipation capability (from the die to the heatsink). Given adequate heatsinking and proper voltage supply rails, 4 of LM3886 can deliver 230+ watts rms into 8 ohms without clipping or engaging the SPIKE protection. While it can get
tricky trying to extrapolate the C/W rating for longer pieces of an extrusion based on ratings for 3" and 6" lengths, a sloppy shortcut can be used. Most extrusions are rated for 3" lengths. Find one rated at 2.8 C/W or better for a 3" length, then use 3" per IC. Arrange the board layout to spread the ICs as far apart as
reasonable rather than mounting them immediately adjacent to one another. Orient the sink for maximum airflow across the finned area, and plan on a fan if the heatsink will be in a confined area.
POWER SUPPLY RAILS
Power supply voltage rails should be set at the minimum that will allow the voltage swing necessary for the desired output power. LM3886 will not swing "rail to ail", the clipping voltage can range from about 4V with+30V rails and a 4 ohm load down to around 3V with +35V rails and an 8 ohm load. The data sheet for the LM3886 uses these values/supply rails, which are pretty optimum for single-chip
applications and for our experimenting have worked well in bridged applications. If the voltage supply rails are too low, the result will be excessive clipping and reduced power output. If the rails are too high, the result will be excessive heating with little or no increase in maximum power output. Thermal shutdown may
engage partially or completely before clipping occurs, but the results will be unpredictable. Although LM3886 is rated to handle supply rails up to +42V, in our experience realistic maximums are +28 to 30V for 4 ohm loads and +35V for 8 ohm loads.
PARALLEL, BRIDGED (BTL), OR BOTH ?
Consider how much power you want to develop and the load impedance of the speakers you will be driving. The LM3886 does a fine job driving 4 ohm or 8 ohm loads but is not intended to drive 2 ohm loads (it will drive 2 ohm loads, but it takes very little output power to hit the maximum output current limit under those
conditions). Remember that in a BTL (bridged) configuration each amp only sees half the load impedance, so using a BTL pair to drive a 4 ohm load should be avoided. The same pair will, however, do a fine job driving an 8 ohm load. Operating 2 or more LM3886s in parallel gives increased output current
capability, but more importantly makes it easier to deal with thermal issues. In theory each chip in the parallel bank can still deliver its full rated power. Since the load is shared, each chip is only providing its share of the overall power,
which means it is producing only its share of the overall heat. A pair of LM3886s pushing 50 watts into 4 ohms will have each chip generating half the power and half the heat, meaning each chip is running cooler and providing less current than it would acting alone. The parallel configuration is good for driving lower
impedance loads (like 2 ohms) or for "beefing up" bridged applications (parallel-bridged).
For driving 8 ohm loads up to around 100 watts, a simple BTL configuration will do nicely. Power supply rails should not exceed + 35V. LM3886 can be driven from either the inverting or non-inverting input, National at one time published drawings showing the bridge being set up by driving the inverting input of one LM3886 and the non-inverting input on the other side of the bridge. We prefer driving the non-inverting input of each side of the bridge, which means we use a dual op amp (TLO72, NE5532 etc)to generate the "mirror image" input signals. This approach does require an additional part, but it keeps the gain precisely the same on both halves of the bridge. Since the dual op amp is essentially a pre-amp, by using a pot for the feedback resistor in the first half of the network it is possible to control relative amplitude of the signal going to the main power section. This can be convenient for coordinating a subwoofer with the main
audio channels, for instance. Refer to the drawing in our AppNote on bridging for the circuit and component values we use.
Results are less satisfactory for bridging 4 ohm loads, even with reduced voltage supply rails. Two LM3886s each driving their own 4 ohm speaker will produce more audio power than they will driving a single 4 ohm load in a bridged configuration. For bridging 4 ohm loads, the parallel-bridged configuration should be used.
Parallel-bridging is nothing more than taking two paralleled banks and using the BTL or bridged configuration. The hard part is setting up the parallel banks, bridging is just like using a single pair of chips in the BTL configuration. When using the parallel-bridged combination to drive 4 ohm loads, attention must be
paid to the voltage supply rails. Do not exceed +30V for best results, and in many applications a slightly lower setting may go a long ways in preventing unexpected shutdown.
ENSURING COOPERATION AMONG THE CHIPS
Refer to the circuit drawing below. The important areas are the two resistors that determine the gain (the feedback resistor and the resistor to ground on the inverting input) and the load-sharing resistor on the output. Matching the parallel circuits exactly is of utmost importance, much like using bi-polar power transistors in parallel. Two approaches can be taken - using 0.1% resistors for setting the gain, or handpicking standard 5% resistors. We normally "handpick" since we have a large number of 5% resistors on hand anyway and since 0.1% tolerance units are relatively expensive, especially in small quantities. Setting the gain with such precision is not necessary, but it is vital that every amplifier circuit in the configuration is matched this closely with all other amp circuits in the configuration. Matching to this degree of precision is 1 ohm per 1000 for
those who "handpick". We usually go with a gain of around 20, if we need more we obtain it "upstream" in either the bridging network or inverting buffer following the summing network (subwoofer amp).
Outputs are joined to the load through 1% 0.1 ohm precision resistors (3 watt minimum), similar to using low value emitter resistors when parallelling bi-polar power transistors. Precision matching is critical here also, to ensure each amp circuit drives the load rather than interacting with other amps sharing the load.
Since a parallel-bridged configuration counts as "extreme conditions", care must also be taken to ensure each amp in the configuration is completely stable and not subject to oscillation. Proper board layout is critical. On the inverting input, we use a 100uF cap (to eliminate DC offset at the output) in series with a 1K
resistor handpicked to match other amp circuits in the configuration. For matching purposes the cap can be ignored since it only has an effect at extremely low frequencies. We also use a 47pF-4700 ohm RC network in parallel with a 20K feedback resistor. Again, for matching purposes the RC leg can be ignored since it only comes into effect at frequencies well above the audio range.
An op amp should be used to drive the LM3886s to prevent overloading the signal source. In theory a bank could be constructed with 3, 4 or more ICs matched in parallel. Practical limits are determined by load impedance and voltage supply limitations presuming thermal isues have been adequately overcome. Developing a Vpp swing of 160V, for instance, cannot happen even in a bridged configuration
powered at +35V. An optimized design will hit the limits for current, voltage and thermal conditions at approximately the same point. Power supply rails, load impedance and input signal levels should be coordinated such that a maximum input signal condition will drive the amplifier to just short of
clipping, which should be reached before thermal or output current limitations are met.
OUR EXPERIMENTAL AMPLIFER SETUP
Our power supply is a switched mode unit regulated at +35V and capable of delivering in excess of 20A. We know it will produce up to 1400 watts, but can't keep enough power coming into it to sustain more than about 800 watts. Since it starts at +12VDC, input currents are pretty large, easily exceeding 100A.
We are running 8 of LM3886. Four of these are "full range" audio each driving their own 8 ohm, 3 way speaker box. The other 4 units are in a parallel-bridge configuration driving an 8 ohm subwoofer. The subwoofer circuitry and first pair of LM3886s are built on one of our LM3886 Experimenter boards, additional pairs are built on boards identical to the power amp sections of the Experimenter board. The
LM3886s are mounted on an 18" length of PEI682 extrusion, the switching power supply is mounted on an identical extrusion. Hearing protection is mandatory when driving the speakers near full power, which we can only do during off hours when surrounding businesses are closed.
Back to App Notes
Back to DC Home Page